More on the semantics of focus and givenness
More on the semantics of focus and givenness
Abstract and Keywords
The previous chapters introduced specific characterizations and even definitions of focussing and similar concepts, postponing their justification to this chapter. The literature contains a huge number of informal and pseudo-formal alternative characterizations, which from a distance may all seem equivalent. Closer scrutiny reveals that most of them are inadequate, or simply untestable, thereby making the case for the concepts used so far. Still, various empirical challenges turn out to be very real, even where the proposed solutions prove to be in need of refinement. Particularly relevant in this connection are observations by Wagner which strongly suggest that focussing requires ‘contrast’ in a sense that is stricter than that advocated by Rooth or Schwarzschild (and previously deemed appropriate, if perhaps not entirely intuitive), discussed in detail here. Other unresolved issues and problems in alternative semantics are reviewed at the end of the chapter.
This chapter has three main parts. Sections 5.1 and 5.2 compare the discourserelated characterizations of focussing and givenness utilized in the previous chapters to possible alternative characterizations, and provide arguments in favor of the former. Readers who are already sold on the discourse-related meanings for focus and givenness may postpone reading these sections until they need to prepare for a discussion with someone who isn’t.
Section 5.3 reviews some recent arguments in favor of strengthening the pragmatic F-relation to include something like True Contrast. It thereby directly extends the discussion in Chapter 3 and points to what I perceive to be genuinely open problems for focus semantics at this time.
Section 5.4 continues by discussing some other, slightly more technical facets of Alternative Semantics and points to a number of open ends, both technically and empirically.
In Section 2.1, I characterized being given as being hyponymous or synonymous with a salient meaning in the context (made salient, for example, by uttering a word with that meaning). Let us now examine this characterization in more detail.
5.1.1 Salient, not familiar
For referring expressions, givenness and FAMILIARITY (in the sense of the familiarity theory of definiteness, e.g. Heim, 1982) boil down to the same effect: If the referent has been introduced by a previous expression, it will be familiar and salient, as in example (1).
However, as already noted in Section 2.1, and illustrated again by examples (2) and (3a), givenness is neither restricted to referring expressions (and thus cannot be (p.100) reduced to familiarity), nor does it coincide with it, even for (putatively) referring expressions.
5.1.2 Salient, not previously mentioned
By assumption, uttering an expression makes its meaning (at least temporarily) salient. Could we not have defined givenness in terms of (recent) mentioning directly, thereby avoiding the somewhat elusive notion of saliency of meaning? We could, but at the cost of excluding certain other examples. Meanings can become salient in ways other than mentioning. If during my visit to your house a dog walks into the room, I could comment as in (4), where the noun ‹dog› is (and has to be) unaccented:
Again, ‘dog’ counts as salient (and hence ‹dog› is given), although it hasn’t been mentioned.1
5.1.3 Salient, not presupposed
In many cases a proposition is salient, because it has previously been asserted, as in example (5). Since an assertion of S routinely results in S’s content being added to the common beliefs of the discourse participants, givenness of a clause in this case coincides with its content being presupposed (note the factive verb ‹know› in (5)).
However, a clause may be deaccented, even if its content is clearly not presupposed, as in (6).
(p.101) And in fact, if we take occurrence under factive verbs as indicative of being presupposed, it is equally easy to find examples where the content of a clause is presupposed, but the clause is accented regularly.
It is not difficult to see what is happening in these examples: Something may well be shared knowledge among the participants without therefore being salient: The guests’ very presence in example (7) is a very good indidication that they did not run out of gas, yet running out of gas is not salient at that point in the conversation, nor is the meaning of ‹gas›, for that matter. Similarly, beliefs that are shared as a matter of world knowledge (that the speaker’s mother is a senator in example (8)), completely unrelated to the discourse situation, do not have to be salient.2
In short, the saliency of a proposition does not entail that that proposition is a shared belief, nor vice versa; this is a first, good reason to keep presupposition and givenness apart.
A second reason is that any lexical category, and all manner of syntactic constituents, may undergo givenness deaccenting, whereas only declarative clauses have the kind of denotation—a proposition—that can be presupposed. Therefore, we couldn’t even coherently say that (the meaning of) deaccented ‹jump› or ‹dogs› in examples (2) and (4) is presupposed.
Of course, we could take the existential closure of those meanings instead, which is a proposition. But this will bring back the previously mentioned problems with a vengeance: a proposition may be mutually believed but not salient, as well as salient but not believed. In each case, it is saliency that appears to be crucial for deaccenting and, by assumption, givenness.
(p.102) For example, intuitively, it is not a mutual belief in example (2) that someone jumped (someone jumping is salient after the first sentence in (2), but it is not assumed that someone jumped—yet). So ‹jump› should not count as ‘presupposed’, and hence should not be given. Alternatively, if we insist that its is generally shared knowledge that someone (somewere, at some point) jumped, then that would predict that the verb ‹jump› could be deaccented; but so should (virtually) any other content word, and completely regardless of context, making the presupposition requirement all but vacuous.
Note, finally, that just as a presupposition theory of clausal givenness would wrongly predict any complement clause to factives to be deaccented, an ∃Clo presupposition theory of sub-clausal givenness would wrongly predict every definite DP to be deaccented, a case in point being definites referring to inferable discourse referents (i.e. referents which have not been introduced themselves, but whose existence and uniqueness can be established in relation to another, already established referent), such as ‹the neck› in example (9):
According to standard wisdom, ‹the neck› in (9) is definite because speaker and addressee know that a guitar has a neck, so that in this context the presupposition of the definite DP—that there is a unique neck—is met.
Yet, ‹(the) neck› must remain accented in (9), despite its ∃Clo being a mutual public belief; this appears to hold in general for such definites, like ‹bus—the driver›, ‹story—the beginning›, ‹band—the drummer›, etc. Like factivity, definiteness and givenness often go together, for obvious reasons, but they are not completely correlated. The reason, I submit, is because different discourse factors underly them: mutual belief versus saliency.
Deaccented elements are also sometimes described as denoting “old information” or being “uninformative.” Presumably, this is intended to mean the same as “presupposed,” and it seems to me an unfortunate choice of terminology, for the same reasons: To be informative is a property of propositions, but not of words or single phrases; and even where an expression does denote a proposition (or a proposition can be derived from its meaning, for example by ∃Clo), the correlation between that proposition being (non-)informative and the expression being given is, as we have seen, imperfect in both directions: That the Johnsons show up would be informative in the context of example (6), but the proposition is already salient, whereas that the guest did not run out of gas in example (7) is not informative, but new in the sense of not salient. Therefore, apart from the slightly oxymoronic flavor of “old information,” “informative,” like its pseudo-antonym “presupposed” seems simply inaccurate as a decription of the pragmatics of deaccenting.
In closing, let me point out that the idea that focussing has an existential presupposition, while clearly related to the “given=presupposed” idea discussed in this section (and, I believe, equally mistaken, see Section 5.2.1), is also different in crucial respects: Just now, we contemplated the idea that the ordinary meaning of a declarative clause or the existential closure of a smaller constituent is presupposed, whereas a focal presupposition theory would assume that the focus closure of a (p.103) clause has to be presupposed. We will return to these issues in Sections 5.4.2 and 5.4.3.
5.2.1 No truth conditions for focussing
It is sometimes suggested that F-domains introduce an existential presupposition, for example ‹KIM took Harry’s book› would presuppose that someone took Harry’s book.3 This idea has been refuted in various ways. First, the sentence can answer the questions in (10), neither of which implies (and hence licenses a presupposition to the effect) that someone took Harry’s book.
Third, there is clear difference between (12a), which—due to the cleft construction—does presuppose that someone took Harry’s book, and (12b)— plain free focus—which does not (see Rooth, 1999: for an extended version of this argument).
Similarly, it is sometimes suggested that free focus entails exhaustiveness of the answer. But again, there is notable difference between plain focus, (13a), and a cleft, (13b), with only the latter truly entailing exhaustivity (and hence creating a contradiction; see again Rooth, 1999).
Data like these show that existence and exhaustivity may be conversational implicatures of focus, but are not parts of truth conditional meaning.
(p.104) 5.2.2 Focus-mentalism
An intuitive characterization of focus meaning would be that focus marks what the speaker intends to highlight, or emphasize, or regards as most important, or most informative (e.g. Miller, 2006). This is an example of what I will call a MENTALIST theory of focus meaning. While such a statement jibes well with our intuitions, and in many case is probably true, too, there is a rather severe problem with using such a mentalist characterization as the meaning of focus: the attitudes it ascribes to speakers are difficult if not impossible to verify independently and systematically. But this is what a theory must allow in order to be testable.
A variant of this problem regards characterizations like “focus presupposes the existence of a closed set of relevant alternatives”:4 if we read this extensionally, then—since we can always find a number of alternatives (i.e. real world entities) for any expression, which exist—this would be trivially met. Understood intensionally, this may mean that discourse participants have to have a number of specific alternatives in mind; but this, again, would be immensely difficult to track down.
More formal-looking characterizations like “focus creates/presupposes an open formula and then provides a unique value for the variable in that formula,”5 or “focus instructs the hearer to open a new file card and write something on it,”6 too, are haunted by basically the same problem: They presuppose the reality of a certain kind of representation in the participants’ minds (assuming in this case that formulas would not be supposed to be objects in the real world), to which independent access would be needed in order to make the claim falsifiable.
For this reason, Chapter 2 followed most formal work on the interpretation of free focus and took felicity in a discourse to be the main data to be accounted for: Which focussing of a given sentence is felicitous as an answer to a particular question, or in response to a particular statement, or as a narrative continuation of a previous text or sentence. The meaning of focus under that view consists of discourse appropriateness conditions. Call these DISCOURSE-RELATED approaches to focus meaning.
(One may argue that on such a view, focus doesn’t have meaning so much as merely a pragmatic function; this is a valid point, though I will continue to use the term “meaning” in what follows.)
At the same time, many if not most discourse-related approaches will—implicitly or explicitly—assume that, for example, Question-Answer Congruence, does not really constrain the relation between a focussed declarative sentence and a preceding interrogative sentence, but between a focussed declarative sentence and some sort of context representation construed on the basis of the question (and maybe additional context as well). This allows the analyst to extend the analysis to cases in which, say, a pertinent question is assumed to be on participants’ minds, but never explicitly uttered: (p.105)
Here we can assume that the question of who has keys to the room is on A and B’s minds, even though it hasn’t been uttered. A plausible analysis assumes that focus in A’s answer is licensed by Question-Answer Congruence, where “question” is understood as the QUESTION UNDER DISCUSSION, QUD, in a discourse model (e.g. Roberts, 1996). Overt question-answer sequences would then be the special case of a QUD set by an actual question utterance.
So even discourse-related approaches to focus are mentalist in this sense. But in contradistinction to approaches that talk about “importance,” “highlighting,” or open formulae in speakers’ minds, discourse-related theories show a direct way in which to access the pertinent mental discourse representation independently: when there is an explicit context (e.g. an interrogative utterance), it has a defined effect on context representation (e.g. setting the QUD); and all relevant aspects of context representation can at least be manipulated by particular utterances such as explicit questions, statements, etc.
That is, the theory, while (at least in part) ultimately about mental objects, can be applied and tested purely as a theory about linguistic data.
Most of what I have said in this subsection applies mutatis mutandis to the “inner workings” of focus theories. Characterizing, for example, Rooth’s (1985, 1992b) theory as asserting that focus “invokes,” “activates,” or “raises” alternatives (in participants’ minds) involves what we may call the PSYCHOLOGIZING of focus semantics.7
While it is perfectly possible that mental “activation,” etc. of alternative meanings takes place, at least in some instances, this is not part of any formal theory of focus I am aware of, nor has it been shown in any psycholingustic experiments. Alternative Semantics uses F-alternatives as part of its technical apparatus, but does not generally claim that F-alternatives become somehow active or salient in the mind of the person producing or comprehending focus (any more than phonology claims that the use of [m] “invokes” nasality). In fact, stating things that way not only does nothing to illuminate the semantics of focus, it also potentially obscures a discriminating view on cases in which it seems plausible that alternatives are made salient, for example out-of-the-blue uses of contrastive accent patterns, as opposed to cases in which they are already salient, or when used in corrections, as opposed to cases in which it is not particularly plausible to suppose that anyone thinks of specific alternatives, such as question-answer pairs or elaboration focus in general.
We now switch from justifying the basic assumptions we have made so far, to highlighting some of their limitations. So far we have assumed, with Rooth (1992b) (p.106) and Schwarzschild (1999), a rather weak F-condition, which invoked no notion of contrast other than “be/entail an F-alternative of” (Section 3.1.3).
The crux, as Rooth (1992b) points out, is that the meaning of the word “contrast” in the definition of CONTRAST in (15) is itself in need of explication; existing attempts in the literature have either fallen short of giving an explicit definition, or provided one which is too narrow to capture all the kinds of cases Rooth intended to be subsumed under his proposal. Moreover, Rooth claims, there is no need for such a stronger notion, since the weakest notion—GIVENNess or Rooth’s original ⊆ / ∈ condition—seems to capture all the cases we need to.
This last point, however, has been challenged by Wagner (2006, 2012b), who makes a rather convincing point for using a stronger F-relation, roughly along the lines of definition (15) above. I will now review the arguments to that effect.
5.3.1 Deaccenting requires local contrast
In the context of (16), (16a), with a deaccented noun, is an acceptable answer, as expected: ‘high-end convertible(s)’ is an F-alternative to ‹cheapF convertible›. By the same token, however, (16b) would be expected to be fine, too, since the F-alternatives of ‹REDF convertible› are the same as those of ‹CHEAPF convertible›. But deaccenting here seems odd. The natural response in the context provided in (16) in this case is (16c).
Büring (2012b) reports similar judgments for a number of sentences informally surveyed with about a dozen speakers, using a questionnaire. In the following two examples, at least half of the speakers preferred deaccenting the noun and accenting the adjective alone:
(p.107) On the other hand, less than a third of the speakers accepted a deaccented noun in the following examples, and nearly everyone preferred a regular accent pattern with accent on both A and N, the latter nuclear:
The contrast between (17) and (18), in particular the lack of deaccenting in (18) raises the same question as Wagner’s (16b)/(16c): Why can the noun, whose meaning is clearly salient in these discourses, not be deaccented as given?
5.3.2 Wagner (2012b)
Intuitively, what goes wrong in example (16b) above is that deaccenting the noun (inappropriately) suggests that ‘red convertible’ truly contrasts (again in an intuitive, non-technical sense) with ‘high-end convertible,’ which it does not. ‘Cheap convertible,’ on the other hand, does contrasts with ‘high-end convertible,’ so deaccenting the N in (16a) is felicitous. This is precisely what Wagner (2006) argues, proposing the constraint in (19):
Setting technical questions aside until Section 5.3.3, let us call an F-alternative that is contrastive in the sense intended here a TRUE ALTERNATIVE (to the focussed element, or its denotation), and the relation between such alternatives TRUE CONTRAST.
According to (19), whether or not something is a true alternative is decided at the level of the focussed element itself; “high end” is a true alternative of ‹cheapF›, but not of ‹redF›. But the informal discussion in Wagner (2006) already makes it clear that we should rather think of this in the larger syntactic context.8 Accordingly, the formulation of Contrast contemplated in (15) assumes that whether or not an F-alternative counts as a true alternative is decided at the level of the F-domain.
Let us look at the focus structure for the examples in (16), representative for all examples in (16)-(18), starting with the deaccenting case (16a) in Figure 5.1(i), and the non-deaccenting case in Figure 5.1(ii).
‹convertible› could be G-marked, along the lines of Section 4.1, in Figure 5.1(i), but the stacked F-domain analysis will be more useful for our expositional purposes). As done occasionally before, the contextually supplied values for the focus variables are indicated in the structure to help comprehension. Assuming that ‘high-end convertible’ and ‘cheap convertible’ are true alternatives, the structure in Figure 5.1(i) meets contrast and maximizes on backgrounded constituents by having ‹convertible› backgrounded.9
The analogous embedded-F-domain structure would not be possible with ‹red› in place of ‹cheap›, since ‘red convertible’ and ‘cheap convertible’ are not true alternatives. Rather, the structure in Figure 5.1(ii) must be chosen, in which the only true alternatives needs to be at the root level. On the downside, ‹convertible› (p.109) in Figure 5.1(ii) is not backgrounded, which is why the structure in Figure 5.1(i) is preferable in the case of ‹cheap›.
This captures the essence of the proposal in Wagner (2012b) (though adapted to our present representational conventions10). The basic argument (the contrast between examples (16a) and (16c)), as well as Wagner’s intuitive explanation for it (the lack of true contrast between ‘red’ and ‘high-end’) seem solid and convincing, and have informed virtually all subsequent work on focus semantics. Arguably, the “anything of the same semantic category is an alternative” view that was virtually universally accepted in the wake of Rooth (1985) is incapable of capturing this contrast, and therefore in need of refinement.
On the other hand, an appropriate reformulation of the F-condition is not easily found, and all current proposals I am aware of raise serious questions of their own. We will turn to these matters momentarily.
Before moving on, however, it is worth pointing out that Wagner’s (2012b) proposal to handle these cases—like our adaptation here—crucially relies on not having stacked F-markers. That is to say, it is not compatible with Schwarzschild’s (1999) idea that every constituent must be GIVEN. To see why, consider the status of the adjective ‹red› in example (16c) (Figure 5.1(ii)): ‹red› is not given (no previous mentioning or other invocation of redness), but it is also not F-marked (for that it would have to be truly contrastive), merely part of an F-marked constituent. Effectively, checking the F-condition only for entire F-domains allows for a third status, other than focussed and backgrounded, namely “part of a focus.” (The present adaption of) Wagner’s proposal makes use of this by requiring foci to be truly contrastive, and backgrounds to be given, whereas elements within a focus may be either; the F-condition does simply not apply to them, because they are not an “active” part of any F-domain. No such category can exist in a system like Schwarzschild (1999) in which the GIVENness/F-Condition is applied to every constituent in the tree, nor one like Selkirk’s (1995b), where every non-F-marked constituent needs to be given.
5.3.3 On the notion of contrast
We have not addressed the question of how to define true contrast, and as I said at the beginning of this section, no entirely satisfactory semantic or pragmatic definition has been forthcoming. Wagner (2006, 2012b) assumes that semantic exclusion is pertinent, and that it is to be looked at at a very local level: a cheap convertible is not a high-end convertible, so ‘cheap’ and ‘high-end’ are true alternatives; but a red convertible may be high-end or cheap, so ‘high-end’ and ‘red’ are not true alternatives.
Embedded under ‹collects›, both ‹cheap› and ‹red› license deaccenting of ‹convertibles› in the context of ‹high-end convertibles›. The question is why (20b) is possible. Obviously, if ‘red’ categorically failed to be a true alternative to ‘high-end’—as condition (19) above implies—(20b) should be as impossible as Wagner’s original ‹red convertible› example (16b) above, regardless of the embedding context.
Katzir (2013) argues that the difference can be seen at the clausal level: while bringing a red convertible and bringing a high-end convertible are not true alternatives, collecting red convertibles and collecting high-end convertibles are. To get a sense of this, compare the rather natural example (21a) to the somewhat odd (21b).
Collecting red convertibles does not exclude collecting high-end convertibles, so Wagner’s (2006, 2012b) characterization of true contrast as semantic incompatibility would still exclude (21a). But, following Katzir’s lead, ‘only collecting red convertibles’ would exclude ‘collecting high-end convertibles’; whereas, even strengthened in this way, ‘only bringing a red convertible’ still does not exclude ‘bringing a high-end convertible.’
Katzir (2013) further formalizes this intuition using the notion of INNOCENT EXCLUSION (a concept he adopts from Fox, 2007). Roughly speaking, an alternative A to a sentence S can be innocently excluded if “S and not A” does not entail any alternative that S alone doesn’t entail already. For example, if we say ‹John brought a redF convertible›, ‘blue’ is innocently excludable: ‘John brought a red convertible and didn’t bring a blue convertible’ doesn’t entail any alternative of the form ‘John brought a x convertible’ (except for ‘John brought a red/a non-blue convertible’ which already follows from ‘John brought a red convertible’).12
However, ‘John brought a red convertible and didn’t bring an expensive convertible’ entails that John brought a non-expensive convertible, which does not follow from him bringing a red convertible. So ‘expensive’ is not innocently excludable here.13
(p.111) In examples with intensional verbs like Katzir’s (20b), however, we can innocently exclude ‘expensive’ on the basis of ‹redF›: ‘John collects red convertibles and does not collect cheap convertibles’ does not entail any additional collection habits on the part of John, in particular, it does not follow that he collects expensive convertibles. Therefore, ‘cheap’—as well as ‘expensive’—are innocently excludable in (20b), and focussing ‹red› is legitimate.
To be sure, ‘John collects red convertibles and does not collect cheap convertibles’ does entail something not entailed by ‘John collects red convertibles,’ namely that he does not collect cheap convertibles. But not collecting cheap convertibles is not an alternative to collecting red convertibles; not collecting something never entails collecting something else, so all of the F-alternatives in such cases are innocently excludable.
To make the following discussion more precise, I will give a formal implementation of this idea in the form of condition (22). Innocent exclusion as defined in Katzir (2013) presupposes that the alternatives in question are all propositional. To still allow our condition to apply to any constituent, condition (22) appeals again to the ∃Clo of the alternatives (recall that the ∃Clo(p) = p if p is already a proposition).
What we now need to convince ourselves of is that examples (16a) and (20b) above meet condition (22), whereas example (16b) does not; structures for these are given in Figures 5.2(i), 5.2(iii), and 5.2(ii), respectively.14
The ∃Clo of C in Figure 5.2(i) is ‘high-end convertibles exist’; from ‘cheap convertibles exist but high-end convertibles don’t’ we can conclude that there are no high-end convertibles, but that, again, is not (the ∃Clo of) an alternative of ‹CHEAPF convertible›, so Figure 5.2(i) is predicted to be fine. ‘Red convertibles exist, but high-end convertibles do not’ entails that non-high-end (red) convertibles exist, which is a F-alternative and which does not follow from ‘red convertibles exist’, so Figure 5.2(ii) is predicted to be unacceptable. Finally ‘someone collects red convertibles, but no-one collects high-end convertibles’ does not entail anything of (p.112)
5.3.4 Where and when is focus contrastive?
Anaphoric deaccenting revisited: Wagner’s (2012b) cases of noun “re-accenting” in A+N combinations are quite striking and justify the introduction of a notion like true contrast, as formally implemented in the definition of . But what determines when a focus is to be interpreted via ∼, and when it is to be interpreted via the stronger Wagner (2012b) explores the most interesting answer to that question, namely that all foci are to be interpreted as truly contrastive (by in our terms). But if that is the case, why has the universally truly contrastive nature of focus not been realized in other instances? How, in particular, could givenness-deaccenting (“anaphoric deaccenting”), as in example (23) have been explicitly argued not to be contrastive (and therefore be different from focussing)?
(p.113) According to Wagner (2012b), the first crucial step in addressing this question is to realize that so-called givenness deaccenting typically happens to immediate constituents of the clause; more precisely, it may occur to those and only those (sub-)constituents of a clause that can be scoped to the clausal level at Logical Form. All the examples that show the stronger true contrast requirement have—Wagner (2012b) argues—the element that would be deaccented inside an island, such as a complex DP.17
The bracketed clause (24a) is then assumed to be the focus projected by the accent on the transitive V. This alone, however, does not yield the right result, at least if we apply condition (22) to (24a), as in Figure 5.3.
‘Smith walked in’ is one of S’s F-alternatives, ‘Smith Q’; but from ‘Smith was arrested and Smith didn’t walk in’, we can conclude other properties (=F-alternatives) that Smith has, for example that he stayed outside, which do not follow from ‘Smith was arrested.’ This informal diagnosis is confirmed if we apply the formal definition of innocent exclusion: ‘Smith walked in’ is not innocently excludable given that Smith was arrested, and the F-alternative set ‘Smith Q.’ This confirms the naive intuition that ‘walking into the store’ and ‘being arrested by a police officer’ are no more truly contrastive than ‘red’ and ‘cheap.’
Wagner’s (2012b) second crucial step, therefore, is to apply a different definition of true contrast to clausal (propositional) F-domains than to all other cases, including that of A+N. According to the propositional definition of true contrast, ‘Smith walked in’ can be innocently excluded in example (23) as long as ‘Smith
(p.114) was arrested by a police officer and nothing else happened/is the case’ entails that Smith did not walk into the store. This is not just true for this case, but for any property P, except those where ‘P(Smith)’ is an entailment of ‘a police officer arrested Smith.’18 Non-clausal constituents, including A+N combinations like ‹red convertible›, on the other hand, follow a different rule in Wagner (2012b): they may contrast with any target A*+N if “everything is an AN” entails “there is no A*N.” This works for the standard cases: ‘everything is a cheap convertible’ entails ‘there is no high-end convertible’ but ‘everything is a red convertible’ does not entail that. Therefore ‘cheap convertible’ but not ‘red convertible’ is a true alternative to ‘highend convertible.’19
Note that if, instead, we applied Wagner’s (2012b) propositional condition to the A+N cases, there, too, virtually anything would go: ‘everything is a convertible that is red and has no other property’ (has no property among the F-alternatives other than those that follow from being red) entails ‘nothing is a blue convertible’, but also ‘nothing is an expensive convertible,’ and vice versa. So if ‘cheap/expensive/blue/red’ are all among the F-alternatives to begin with, any pairing of these is predicted to be truly contrastive. Put differently, Wagner’s Contrast Requirement for propositional F-domains, unlike that for non-propositional cases, is, intuitively, not at all contrastive. There are practically no cases it excludes.
In sum, Wagner (2012b) proposes a classificatory diagnostic for when true contrast (in the Katzir, or non-propositional Wagner sense) is required for deaccenting—only for things that cannot move to a clausal level at Logical Form. But crucially then, two very different F-conditions (∼ and in present terms), are applied to the two configurations. There is, as far as I can see, nothing about (p.115) the syntactic environments which define the two classes from which the difference in F-conditions follows. The association “propositional/moveable → ~, non-propositional/non-moveable has to be stipulated. This holds regardless of whether is spelled out as in Wagner’s work (which, however, wrongly rules out cases like Katzir (2013: (20)) or as in Kazir’s.
Answer-focus revisited: Apart from standard cases of anaphoric deaccenting, Contrast as defined in (22) also runs into problems with plain answer focus.
Intuitively, ‘who (arrested Smith)’ and ‘Jones (arrested Smith)’ are not (truly) contrastive. Likewise, formally, the ∃Clo of the question is ‘someone arrested Smith’ (see Section 3.5.2), but ‘Jones arrested Smith and no one arrested Smith’ is contradictory, so ‘someone arrested Smith’ cannot be innocently excludable. This problem applies generally to cases of what we called ELABORATION Focus, including examples (26) and (27).20
A tempting response to this problem is to claim that the answer in such cases is indeed truly contrastive, only it does not contrast with the question, but with other potential answers. This line of agument is made explicit, for example, in Wagner (2012b):
The context [in example (25); DB] makes available a set of propositions of the form ‹x arrested Smith› … [T]he stress shift can be explained in the same way as we would explain it if the context included the statement ‹Sally arrested Smith›. (Wagner, 2012b: sec. 1.2.6)
Schematically, the argument goes like this:
Now, note first that, if the same line of argument is to apply to examples (26) and (27), then we have to assume that either these denote sets of propositions, too (‘that Sandy arrested Smith’ ‘that Charles arrested Smith’ …; ‘that we bring French wine’ ‘that we bring German wine’ …), or they otherwise “make available” the corresponding questions, which in turn make available all their possible answers.21
(p.116) Second, and more importantly, there is a problematic equivocation in (28). The conclusion should be that a question makes salient the set of its answers, not each element of that set. And for a set of propositions to be salient/“available” is not the same as for the propositions in that set to be salient/“available” For example, an utterance of ‹Sally arrested Smith› makes available the proposition that Sally arrested Smith, and hence allows for deaccenting of a clause expressing that proposition, as one would expect, cf. (29). But the same is not even remotely possible in the other two contexts, (30a) and (31a).
Indeed, as the (b) answers show, ‹Who/someone arrested Smith› does allow deaccenting of ‹arrested Smith›, which we take to mean that it makes salient the property of arresting Smith, but not the proposition that Sally arrested Smith.
This exact pattern follows immediately if we assume, as we did before, that utterances in general make salient their ∃Clo, which is ‘that Sally arrested Smith’ for example (29), and ‘someone arrested Smith’ for examples (30) and (31), and nothing more (or alternatively: that making a set of propositions salient has the same effect as making their grand union salient).
I take this to mean that the idea that elaboration focus, including answer-focus, is a “covert” type of contrastive focus is not a fruitful one. At the very least, “making a proposition available” in the sense of the quote from Wagner (2012b) would have to be a rather different, weaker, notion from “making a proposition salient” in the sense used throughout this book so far. The situation that would—hypothetically— result is schematized in (32):
(Certainly this is not Wagner’s (2012b) position, which set out to argue that givenness deaccenting and answer-focus are all the same phenomenon.)
In sum, imposing a true contrast requirement on focussing, even if properly worked out, works well in A+N cases like Wagner’s original examples and Katzir’s extended data set, but it wrongly predicts that all instances of deaccenting are actually—and contrary to all traditional belief—truly contrastive in the same way. There is as yet no uniform condition that could be successfully applied to all cases.
(p.117) 5.4 Open ends in Alternative Semantics
5.4.1 Focus on semantic functions
Up to now, when we paraphrase the F-alternatives to a VP like structure (33), we use formulations like ‘do something to Kim’ or ‘stand in some relation to Kim’.
These characterizations, intuitively adequate though they may be, are not borne out by our formal semantics, however. In fact, shockingly, the formalism predicts that the F-alternatives of (33) are the same as those of (34) (with or without stacked F-markers on V and DP).
How so? The alternative set of (34) is the set of all properties. A transitive verb like ‹kiss› alone, on the other hand, denotes a function from individuals to properties (also called a RELATION), and so its alternative set is the set of all relations. Now— and this is where intuition and formalism part ways—for any property P, we can define a relation, call it RP which maps Kim, or every individual at all, for that matter, onto P (we could write this function as λx ∈ E.P).
A fortiori, then, for any property P ∈ 〚 [kiss Kim]F 〛 F, there is RP (which, being a relation, of course is in 〚 kissF 〛F, too) s.t. RP(Kim) = P, and consequently 〚 [kiss Kim]F 〛F = 〚 kissF Kim 〛F.22
So any arbitrary property, say, to be drinking beer, is an element of 〚 kissF Kim 〛F, because we can find, among the F-alternatives to ‹kissF›, functions which map Kim to the property of drinking beer. Generally, the following holds:
One suspects that this is not a good thing, and indeed we can find cases in which this creates very wrong predictions. For example, (36-i) is predicted to be a good reply on Ali’s part to convey what intuitively only (36-ii) can: That Captain Hook told Ali, not Ms Summerset, where the treasure is buried.
(p.118) To see how, pick as an F-alternative to ‹TOLDF› that function φ which maps Ali (the speaker) to the property π of telling Ms Summerset where the treasure is buried. Since π is made salient by Al’s utterance, Ali’s reply (36-i) should count as a contrastive focus sentence (the problem cannot be the lack of accent on ‹me›, as answer (36-iii) shows).
The problem originates with functions like φ (or the one that maps Kim to the property of drinking beer) being among the F-alternatives of ‹kissF› and ‹tellF›. If the F-alternatives were restricted to the kind of denotations that natural language words can express, the problem would presumably go away.
Embarassingly enough, though, it is not straightforward to define even what a “natural” natural language meaning is. In other words, not only do we allow φ as a focus alternative, we cannot even block it as the meaning for a transitive verb.
One can conceive of several directions in which to look for a solution to this problem. One would be to grab the bull by the horns and develop semantic restrictions on the notion “natural word meaning,” for example by way of meaning postulates, which somehow enforce that the meaning of a transitive VP depends in an intuitive way on the properties of the object referent. One could then enforce the same restriction on F-alternatives. Essentially F-alternatives would be restricted to the meanings of existing (or randomly missing but possible) words or phrases in the language.
Another would be to assume that something structural restricts F-alternatives. Assume, for example, that ‹kiss Kim› means something like ‘do something to Kim which is a kissing,’ and 〚 kissF Kim 〛F is really ‘do something to Kim which is an X-ing’ or ‘be in relation X to Kim which affects Kim.’ That is, assume that certain parts of the verbal meaning are not encoded by the F-marked expression itself (or at any rate not affected by it), and therefore invariant across alternatives.
This route is pursued—for very similar reasons—in Bonomi and Casalegno (1993), which models sentence meanings as event description. Thus the F-alternatives for ‹Sam KISSEDF Kim› (called the BACKGROUND MEANING in Bonomi and Casalegno, 1993) would literally be ‘there was an event (of unspecified type) of which Sam is the Agent and Kim is the Theme.’ Under reasonable assumptions about what it means to be a Theme of an event, a beer drinking would probably not qualify (and even if it did, it would still have to be a beer drinking of which Kim is the Theme—whatever that would mean—so at least Kim would have to be salient as well).
The same kind of solution is possible in standard Alternative Semantics as well. Either by adopting the event-based semantics wholesale, or by restricting F-alternatives of transitive verb meanings to relations that entail “agent-hood” and “theme-hood” of their respective arguments in some other way. Since not much hinges on the details of implementation I will not explicate this any further here.
One way or the other, it is probably a good idea to restrict F-alternatives to ‘natural’ meanings, either at the level of introducing them (that is, restrict 〚 EF 〛F) or at the level of retrieving them (restrict 〚 E 〛Fin the configuation ‹[E ∼Ci ]›).
(p.119) 5.4.2 Givenness distributivity
A question I have not seen discussed in the literature is the one in (37).
On the face of it one suspects that the answer is “yes.” After all, how could, say ‹eat beans› be given, without both ‹eat› and ‹beans› being given? Yet there are constructions in which things are not that easy.
Example (38a) reminds us that contrastive focus can work via entailment/hyper-nymy: ‘someone is learning the piano’ entails ‘someone is (and defeasably: wants to be) learning a musical instrument’.
So in (38b), since ‘someone refusing to learn any musical instrument’ arguably entails ‘someone not practicing piano’, we might expect that ‹won’t practice the piano› in (ii) is given and could therefore be deaccented. But the judgment is that deaccenting the verb phrase as in example (38b-i) is odd, and that, in particular, ‹piano› should bear a PA, as in (38b-ii).23
The problem with (38b) is likely not the inference from ‘refuse to learn any musical instrument’ to ‘not practice’, because precisely that inference is needed to explain the felicitous deaccenting in example (38b-iii).
One does not have to look too far for an intuitive explanation for the contrast between (38a) and (38b-i). While ‹won’t practice piano› might be given in the context of (38b), ‹piano› itself is not, and therefore cannot be deaccented. This is different in (38a): not only is ‹learn a musical instrument› given (since ‘learning the piano’ is salient), but so, too, is ‹musical instrument›, being a hypernym of the (p.120) salient ‘piano’; likewise in (38b-iii), both ‹won’t practice› and ‹practice› are given (by ‘refuse to learn any musical instrument’).
So example (38b-i)/(38b-ii) appears to be the crucial case: a (technically) given constituent—‹won’t practice the piano›—contains a non-given constituent— ‹piano›. And the accenting facts suggest that in this case the given constituent nevertheless needs to contain an accent, in fact the NPA.
How do the various theories of F-interpretation measure up to this kind of case? The accenting on ‹piano› is predicted by a theory like Schwarzschild (1999), or Selkirk (1995b): Since every constituent needs to be checked for GIVENness (or givenness, respectively), F-marking on ‹piano› in example (38b) is obligatory— regardless of the givenness status of any constituent containing it. It is, however, unexpected given systems like Rooth (1992b), Wagner (2006, 2012b) or Büring (2012b), which do not use stacked Fs as newness markers and represent example (38b-i)/(38b-ii) above as in Figure 5.4.
As long as ‘someone won’t practice the piano’ is entailed by something contextually salient, these structures should be fine. Indeed, any type of structure that would instead mark ‹piano› as an additional focus, such as those in Figure 5.5, might not meet the F-Condition: ‘that someone practices something’ is not entailed by anything salient in (38b), unless we take ‹learn any musical instrument›, without the ‹refused to› part, as a possible antecedent.
Even more clearly, ‹practice piano› does not meet any stronger condition on truly contrastive F-domains: no property ‘practice x is salient in the context. Finally, it is far from obvious that the structures in Figure 5.5 have more backgrounded elements than (i) in Figure 5.4, rather than fewer (‹piano› is in the background in Figure 5.4(i), but not Figure 5.5(i))), or the same (‹practice› is already contained within a G-marked constituent in Figure 5.4(ii); G-marking itself, as in Figure 5.5(ii) hardly makes it more backgrounded).
So adopting the stacked-F assumption—effectively that every constituent is an F-domain—would correctly predict the lack of deaccenting in these downward entailing contexts. But recall from Section 5.3.2 that any true contrast requirement on foci is incompatible with stacked-F systems (since these do not allow for a constituent to be neither given nor F-marked).
What can be done to get the best of both worlds? Since this question has not, to the best of my knowledge, been discussed in the literature, I can only offer some speculations.
(p.121) One route to explore would involve a refinement of the notion of saliency; effectively a salient meaning μ need not make all its entailments salient, or maybe at least not in DE contexts. The effect to be gained is that while ‘playing the piano’ makes ‘playing a musical instrument’ salient, ‘not practicing’ would not make ‘not practicing the piano’ salient (even though it entails it). This sounds plausible enough, but it is not clear to me that an elaboration of the idea would not, in the end, involve a “saliency check” on every “part” of a salient property, much like (but more obscure than) the Schwarzschildian GIVENness check on every syntactic constituent expressing it.
On the other hand, there may be independent reasons for wanting to adjust what exactly is made salient by an utterance. In some cases, it appears that an utterance makes salient something it does not entail, as in example (39) below, due to Mats Rooth.24
The verb ‹find› in example (39) is deaccented, even though it is not given: someone looking for something does not entail someone finding something (unfortunately). Yet we might want to say that ‘looking for something’ makes ‘finding something’ salient, resulting in givenness of ‹find›.
Now, might example (39) not be precisely the case we were looking for earlier, namely a constituent being given, and hence deaccented, while containing nongiven material? Note that the representation in Figure 5.6 would indeed license this pattern, due to the fact that ‘looking for a horse’ arguably does entail ‘trying to find a horse’ (this, in fact, was Rooth’s point with this example.)
This would be in marked contrast to our earlier example (38b), where this kind of structure was—judging by the need to accent ‹piano›—impossible (compare Figure 5.4 to the allegedly legitimate Figure 5.6).
Crucially, there is no ‘try to’ in this example. Hence, it is unclear what could license a F-domain here (cf. Figure 5.7), given that ‘looking for the container’ clearly does not entail ‘find the container.’
Again it seems intuitively plausible to think that the context sentence in example (40) makes the property of finding the missing container salient, and thereby licenses givenness and deaccenting of ‹found it›. But there is, in all likelihood, no constituent in that context sentence whose ∃Clo would entail that someone found the missing container, so none of the theories reviewed actually predicts this to be possible.
Where does this leave us? It seems to me that examples like (40) (and maybe (39)) suggest a refinement of what it means to make a meaning salient, since the current definition undergenerates in these cases (not predicting deaccenting where it should). A successful refinement of that notion might at the same time fix some overgeneration problems that arise with Rooth/Büring/Wagner-type theories in connection with downward entailing contexts such as example (38b), but that is a mere speculation. If not, these examples remain problematic, unless we return to Schwarzschild’s (1999) idea that every node is an F-domain, which seems successful in these cases, but problematic elsewhere (see again Section 5.3).
5.4.3 Focus/background compared to new/given once more
Generally speaking, the theories of focussing reviewed so far regularly entail, implicitly or explicitly, the conjecture in (41).
licensing conditions in these theories directly entail that F-less elements must be given (due to the fact that the conditions apply to every constituent in a tree).
We discussed two cases in which, potentially, conjecture (41) may not hold in systems in which not every constituent, but only F-domains have to meet a licensing condition,—such as Rooth (1992b)—in Sections 5.4.1 and 5.4.2 above. I argued that the empirical picture there also seems to support (41), even where technically we might predict otherwise.
What would a bona fide counter-example to conjecture (41) look like? There are two scenarios to be distinguished. First, that a non-given element remains unaccented because it is in the background of a focus. And second that an element is accented because it is non-given, but yet can be shown not to contribute alternatives to some higher focus domain.
According to Neeleman and Szendroi (2004), ‹(to some) kid› in example (42) is unaccented because it is in the background of a contrastive focus, ‹Superman›. The example’s focus structure is argued to be along the lines of Figure 5.8(i).
The gist of Neeleman and Szendrői’s (2004) analysis is expressed as follows:
The discourse status of Mother’s reply is complex, as it contains a contrastive focus inside a contrastive focus inside an all-focus sentence. First, her answer tells Father what happened and therefore … [TP] must be in focus as a whole. Second, the VP ‹reading Superman to some kid› is contrasted with doing his homework. Finally, the DP ‹Superman› is contrasted with decent books. These three foci share a single phonological marking, namely, the stress on Superman. (Neeleman and Szendroi, 2004: 150) (p.124)
The tree in Figure 5.8(ii) fleshes this out a little for perspicuity, using annotated F-variables.25 The crucial point is that ‹to some kid› is not given (and neither is, consequently, the background of the lowest F-domain, VP), yet it is deaccented. In a slogan:
(BON) ‘Backgrounding overrides newness’
It is not clear whether Neeleman and Szendroi (2004) would subscribe to BON in general, or only for cases in which the new-yet-backgrounded material occurs within a larger Focus. The more detailed proposal in Féry and Samek-Lodovici (2006: esp. 137ff), which partly builds on Neeleman and Szendroi (2004), clearly entails the more general prediction that, at least within a limited prosodic domain, contrastive focus may generally license deaccenting of new elements. Apart from the original Superman sentence (42) above, they claim Rooth’s (1992b) farmer example (‹An AMERican farmer was talking to a CaNAdianfarmer.›; see Section 3.1 above) as an instance of BON, as well as examples involving Right Node Raising such as (43) (their (43)), modeled after parallel German examples discussed in detail in Féry and Hartmann (2005).
Are we to conclude from these examples that conjecture (41) above (“elements in the background are given”) is simply wrong and BON is the more accurate generalization? I would caution against this. It still seems rather clear to me that in general, neither free nor associated contrastive foci can license deaccenting of non-given material in their background; witness the examples in (44) and (45).
In all these examples I tried to make a contrast as plausible as possible without at the same time having a parallel F-domain (complete with symmetrical deaccenting). The results still are strongly marked.
A similar diagnosis is, I think, warranted for answer focus. Focussing ‹candle holder› in (46)—which is the predicted answer focus—does not license deaccenting of the new adverbial following it.
In all these cases an (additional) pitch accent on the final NP seems necessary, as would be expected given conjecture (41) above.
So it seems to me that the examples discussed in Féry and Samek-Lodovici (2006) are restricted to a narrowly circumscribed class of constructions, quite likely something like symmetrical deaccenting, though of course more empirical work is required to establish whether that is the correct characterization.
The Superman cases clearly do not involve any kind of symmetry or parallelism, but here, too, I would hesitate to draw any radical conclusions before a more comprehensive empirical picture is available. I think structurally parallel cases to Neeleman and Szendrői’s (2004) do not in general allow backgrounding of a new element. Some rather marked examples are given in (47) and (48).
(p.126) Perhaps it is significant that the deaccented item in the original Superman sentence (42) above was the noun ‹kid› (rather than ‹woman›, ‹bathroom› or ‹uncle› in (47) and (48)); the context does contain the word ‹children›, and clearly the speaker’s kid (and the fact that he is a kid) is salient in it.
In sum, although the empirical evidence is not unequivocal, I will assume for the time being that conjecture (41) above describes the rule, rather than the exception.
Accented but backgrounded: The other case outlined at the beginning of this section—something in the background of a focus that is new and accented—is argued to be found in examples like (49), by Katz and Selkirk (2011).
‹Only› in example (49) associates with the focus ‹Modigliani›; its F-domain is, presumably, VP, including the PP ‹to MoMA›, which is not given. Katz and Selkirk’s (2011) first finding, from a series of production experiments, is that the PP ‹to MoMA› (which is not the focus of ‹only›) in this case is produced with a clear pitch accent. Their second finding is that the pitch accent on the contrastive focus, ‹Modigliani› in example (49), is slightly but systematically stronger than that on the following non-given phrase, ‹MoMA› in (49). The implications of the latter finding will be discussed in Chapter 6. What we are interested in now are the implications of the first finding for focus interpretation. The most straightforward conclusion seems to be that in example (49) ‹only› associates with the focus ‹Modigliani› within the F-domain ‹offer that Modigliani to MoMA›. Anticipating some representational devices from Chapter 10, this can be represented as in the tree in Figure 5.9.
find (in fact, the tree in Figure 5.9 would represent a kind of Superman sentence, discussed at the start of this subsection, in which a new element would be backgrounded and deaccented; indirectly, then, Katz and Selkirk’s (2011) findings provide an argument against the general possibiliy of such sentences).
It is important at this point to realize that the value of the variable C’ restricting ‹only›, ‘offered x to MoMA,’ is not contextually given, since at least the MoMA has, by assumption, not been discussed or otherwise been salient. While this offers an intuitive reason for ‹to MoMA›’s need to be accented, it also raises a red flag about having ‘offered x to MoMA’ being the value of the F-variable ∼C’ in the tree in Figure 5.9, and indeed any F-variable at all: a non-anaphoric F-variable wrongly allows deaccenting of non-given material. Put differently, our choices at this point seem to be: either assume that the variable on ‹only› need not be coindexed with a F-variable, or assume that focussing alone does not license deaccenting. These options are represented in Figures 5.10(i) and 5.10(ii).
The value of C in Figure 5.10(i) could still be constrained by the F-alternative value of its sister, for example it may have to be a subset of it. In that way, we would not lose all connection between ‹only› and prosodic focussing.26 The tree in Figure 5.10(i) also would not be a case of overfocussing, since no further F-domains within VP could be anaphorically licensed, so this is in fact the minimal focussing compatible with the standard F-conditions. The main drawback of this representation is that it does not account for any special prosodic marking on ‹Modigliani›, since it is, on this account, just part of a regular VP-Focus; so if Katz and Selkirk’s (2011) experimental findings turn out to be robust, Figure 5.10(i) does not provide a satisfactory representation to encode them.
The tree in Figure 5.10(ii), on the other hand, distinguishes between the focus ‹the Modigliani›, any given parts (here: only the subject), and the rest, which is (p.128) backgrounded, but not given. This kind of system reduces the function and effects of focussing to a bare minimum: Given foci (see Section 2.4.2), and “extra strong accenting” in cases like Katz and Selkirk’s (2011). All deaccenting is accomplished by givenness alone. This necessitates a complete re-thinking of the pragmatics of focussing (and the question what forces G-making), and though the general lines of such an approach seem clear at this point, I will leave elaboration of the details to a later occasion.
5.4.4 The role of context and world knowledge
A famous example for contrastive focus goes back to Lakoff (1968):
The contrastive focussing of ‹he› and ‹her› is predicted to be possible if there is a salient target (antecedent) of the form ‘x insulted y,’ for example that she insulted him. But all the context provides is the proposition that she called him a Republican. Of course, as Lakoff already notes, the very accenting in example (50) conveys that calling someone a Republican does indeed imply insulting them. Put differently, the felicity of contrastive focussing in example (50) hinges on a presupposition which licenses the inference from ‘she called him a Republican’ to ‘she insulted him’ (which can then serve as the focus antecedent for ‹HE insulted HER).
It is because of examples like these that the licensing relation for givenness and focussing is often taken to be, not logical entailment or strict hypernymy/synonymy, but something like CONTEXTUAL ENTAILMENT, CONTEXTUAL IMPLICATION, or ABDUCTIVE IMPLICATION. The obvious idea is that world knowledge or beliefs, both encyclopedic and probabilistic, can provide additional premises on which to base the conclusion from the ∃Clo of a contextually salient meaning to the ∃F-Clo of an utterance. For example, that she called him a Republican, taken together with the unspoken premise that the allegation of Republican allegiances constitutes an insult, logically entails that she insulted him.
The concept ‹New Yorker› can be deaccented as given in example (51) because the individual Woody Allen is made salient by A’s utterance, which in turn makes New York and its denizens salient by our encyclopedic knowledge that the famous writer and director is a New Yorker.28
A tricky detail not usually discussed in this connection is that it is not straightforward to implement this intuition in terms of privative givenness. Suppose we tried something like definition (52).
The reason (52) is inadequate, in fact dangerous, is that by this condition, anything entailed by shared assumptions of the participants alone would also qualify as given. To give a random illustration, assuming that Ali and Bern know that London exists, (52) predicts that they could deaccent ‹London› at any point, regardless of context.
Formally, since the background assumptions include that London exists, the salient proposition that Bern should come over for dinner together with the background assumption that London exists entails that London exists, which would make ‹London› given (shared (but not salient) assumptions are printed in gray).
As a first step to solving this conundrum we notice that in (54), givenness is actually met by the shared assumptions alone; the contextually salient meaning has no role to play in it. So maybe condition (55) below would fare better?
This works fine for example (54), but it also wrongly rules out example (51) above, as shown in (56): that Woody Allen is a New Yorker all by itself entails that New Yorkers exist (=∃Clo(New Yorker); similarly for example (50) above).
Something like condition (55) would give the correct predictions at the level of the F-domain. So instead of applying condition (55) above to single words and phrases, we should enforce definition (57).
Now we get the correct results: Licensing the larger F-domains requires both a salient antecedent and an additional premise out of the shared assumptions (gray).29 What these cases show again, then, is that givenness alone, at least when bridged by contextual assumptions, is not sufficient to license deaccenting. The (p.131) resulting pattern also has to “fit” in the larger domain. It seems that deaccenting always also involves focussing.
5.5 Chapter summary and outlook
In this chapter I first collected arguments in favor of a strictly discourse-related, anaphoric theory of focus and givenness (Sections 5.1 and 5.2), not least because of the lack of sufficiently precise, workable alternatives. I then turned to what may be described as a laundry list of loose ends for such a theory. Naturally, it is difficult to try to draw any conclusions from that part. In particular, while some of the observations and generalizations seem to favor a reduction of privative givenness to focussing—Sections 5.3 and 5.4.4—others seemed to argue for an even stronger dissociation of focussing and givenness than assumed by standard theories—Sections 5.4.2 and 5.4.3.
Likewise, the idea that the F-condition should include some stronger notion of contrast (which, interestingly, hinges on the idea that deaccenting always involves focussing), while backed up by compelling evidence in some cases (Section 5.3.1), seems to be unsupported in others (Section 5.3.4) (which, incidentally, look like privative deaccenting).
In addition to the phenomena discussed in this chapter, there are at least two types of focussing configurations that are different enough from ordinary focussing (answer, contrast, elaboration) to merit separate mentioning. Since what we know about them at this point does not shed much new light on focus in general, some pointers to the relevant literature will suffice here.
First, so-called POLARITY or VERUM Focus, often realized by a single pitch accent on the finite verb as in (61).
Generally speaking, polarity focus is used where the propositional content of a sentence is given in its entirety, whereas the fact that it holds true is contrasted with someone’s belief or impression that it is not (it is thus the posititve counter-part of narrow focus on ‹not›). For thorough discussion see the seminal Höhle (1992), as well as Gutzmann and Miró (2011), Stommel (2011) and the papers collected in Blühdorn and Lohnstein (2012).
(p.132) The striking thing about (62) is that the focussed element is neither a word nor a morpheme, but simply a prosodic foot, ‹mite›. Artstein (2004) convincingly shows that this kind of WORD-INTERNAL Focus has all the properties of regular contrastive focus, including the ability to associate with ‹only›, and argues that speakers devise an ad hoc interpretation for ‹stalag/stalak›, as a function that maps the expression ‹mite› to the meaning ‘Stalagmite’ (and the expression ‹tite› to the meaning ‘Stalagtite’), and provides a compositional semantics to integrate this into Roothian Alternative Semantics (see Artstein, 2004: for details).
(1) Many indexical expressions, including all first and second person pronouns, as well as things like ‹here, today, now› etc. strongly tend to be unaccented. If this is an effect of givenness, it suggests that speaker, addressee, speech time and place, etc. are always salient, and hence given, even if they haven’t been explicitly mentioned before.
It is possible, however, that these elements are unaccented for other reasons, for example because they are functional, rather than lexical expressions. In that case, no argument against the importance of mentioning is forthcoming from these cases.
(2) Kallulli (2006, 2009, 2010) observes that non-factive predicates like English ‹believe› (and its German and Albanian counterparts ‹glauben› and ‹besoj›) when occurring with a pleonastic object ‹it›/‹es›, or a doubling clitic ‹e›, respectively, take on a factive meaning, and typically—and unlike their ‘plain’ counterparts—occur with deaccented complement clauses:
It seems plausible to think that in this case, factivity and givenness do go hand in hand, perhaps due to the anaphoric or topical nature of the pronoun-doubling construction. However, lexically factive verbs like ‹know›, or emotive-factives like ‹regret› or ‹be happy›, do not show this behavior, as examples like examples (7) and (8) in the main text clearly show, so Kallulli’s (2006: 216) claim that “[i]n order to get a factive reading, the (factive) verb must carry nuclear pitch accent” is likely to only be accurate for the non-factives.
(4) VallduvÍ and Vilkuna (1998: 83).
(7) A simple web search will find dozens examples of this.
(8) Wagner’s example is: ‘used’ is a true alternative to ‘new’ in the context of ‘car’ but not ‘boy friend.’
This would make no difference to the intonation (since only Fs are relevant to that). I assume, however, that the NP ‹convertible› does not count as within a focus domain if it is identical to the focus domain, and hence does not count as an anaphoric constituent, so that the structure in Figure 5.1(ii) is the official representation of (16c).
(10) Wagner’s own implementation differs from the present one in various other ways, most notably in that grammar marks G(ivenness), rather than F(ocus). The central condition on the interpretation of G-marking is paraphrased in (i):
((i)) A structure of the form ‹[A BG]› is well-formed only if there is a (contrasting) alternative A* to A s.t. ‘A* B’ is salient.
(12) To be sure, it is not claimed that e.g. ‘John brought a red convertible and did not bring a blue convertible’ will actually be implied by this sentence (after all, he may have brought two convertibles). Rather this is just a step in determining whether ‘blue’ and ‘red’ are true alternatives in this context, which, in a manner of speaking, does not itself leave any marks on the interpretation of the sentence.
(13) The formal definition of innocently excludable is:
In words: if you start combining p with as many negated alternatives from A as consistently possible, then iff every way of doing so includes ¬a, a is innocently excludable.
(15) In particular, it does not follow that someone collects red cheap convertibles, because the collector of red convertibles need not care about the price of their items at all.
(16) Katzir (2013: sec. 2.3) actually applies his F-condition to the entire clause ‹He brought her a red/cheap convertible›, rather than having the DP be a subordinate F-domain; the resulting predictions are the same. However, the entire clause couldn’t be the F-domain for the focus on ‹cheap› in example (16a)/Figure 5.1(i) according to present assumptions, because of the first clause of the F-Condition: There is no salient meaning of the form ‘he brought a z convertible. (Katzir, 2013 calls the proposition ‘Mary’s uncle brought a high-end convertible’ an “accommodated expectation,” which he seems to assume is sufficient to antecede the focus; we have, however, been assuming that saliency is a more restrictive notion than “expectation”; furthemore, it is unclear in what sense e.g. in Katzir’s original example (20) the assertion that Mary collects high-end convertibles should lead to an expectation that John, too, collects high-end convertibles.)
(17) I will not attempt to assess the empirical accuracy of Wagner’s classification, because in the end, as we will see, the syntactic difference underlying the classification has very little to do with explaining the contrast.
(Wagner, 2012b: 138ff) presents a number of arguments to argue the correctness of his generalization, of which a huge class—the ones involving coordination—seem based on empirically inconclusive data.
(18) The definitions in Wagner (2012b) are not entirely coherent, but the above is almost certainly what is intended, as is evident from the discussion. Wagner’s definition of the exhaustivity operator, his (47), contains an unbound symbol C, which is likely intended to refer to the F-alternatives of the focus to be interpreted. In the definition of his true contrast requirement for propositional constituents—his (48a)—a (likewise unbound) symbol Δ occurs, which should likely refer to the (or a) F-alternatives of the F-domian. For concreteness, I assume the intended definition to be equivalent to (i) (where ⊕ stands for the standard semantic composition operation applied to meaning in the domains of b andf):
The details of what exactly would not be innocently excludable remain unclear, since Exh as defined does not apply in the intended way to e.g. generalized quantifiers; but one way or another it will be a rather small class; consistent with that Wagner (2012b) does not contain any examples of sentential F-domains that are excluded by the condition proposed there.
(19) Curiously, this definition also renders ‘cheap convertible’ and ‘(cheap) bicycle’ truly contrastive (if everything is a cheap convertible, then there are no bicycles); evidently, there must be additional conditions on what may be a contrasting antecedent.
The condition plainly fails in many other cases, among them (i).
‘Everyone knows Kim’ does not entail that no one dislikes Kim, so this should not be a case of true contrast. Of course, the background in (i) is a direct object, and hence moveable; but I do not see how any permitted form of raising either occurrence of ‹Kim› would produce a form to which the propositional rule could successfully apply.
(22) The missing half of this proof is straightforward: For any R ∈ 〚kissF〛F, R(Kim) ∈ 〚 [kiss Kim] F 〛 F.
(23) Other cases in which a given constituent might potentially contain non-given material involve disjunction and tautologies, as illustrated below:
In these cases, too, the (b) sentence is a logical entailment of the salient proposition in (a), and hence predicted to be given. While this seems implausible, I found it difficult to construct pragmatically plausible examples of this kind which would actually demonstrate that this is the wrong prediction, i.e. that disjuncts and tautologies cannot be deaccented without antecedents.
(24) p.c. June 9, 2012.
(25) Neeleman and Szendroi (2004) utilize neither F-marks nor F-domains, so the choice of F-domains in particular may not be precisely what they have in mind, but nothing should hinge on this as long as ‹to some kid› is in the background of some focus. Neeleman and Szendroi (2004) present examples like this one as an argument against using F-marking in general, but as far as I can see, they are only problematic for theories that move foci to a designated clausal position.
(26) It is tempting to think that this is already implemented, since the value of an F-Variable ∼C is allowed to be a subset of the F-alternatives of its sister. The difference is that it still also has to be
anaphoric, lest focussing itself (not just the restriction of ‹only›) lose any relation to context. What we are talking about here is more along the lines of (i):
(28) Alternatively one could hypothesize that New York does not gain bona fide saliency, but that the mere saliency of Mr Allen satisfies the licensing condition on focussing (i.e. the licensing condition itself is not just entailment but something like “entailment cum background knowledge”). Since I do not see any empirical consequences of the choice here, I will continue to equivocate somewhat on this point.
(29) I still include the ‘but not alone’ clause because, for example, it may be part of the shared assumptions that (once, somewhere) someone insulted someone, but we do not want that to be sufficient to get the ‹SHE insulted HIM› prosody (without a salient antecedent about some kind of insult).
(30) From the New Yorker (April 14, 1956, p.36).