Research on Language and Computation 00: 1–18, 2006. c 2006 Springer Science+Business Media, Inc. Manufactured in The Netherlands. 1 On Appositives and Dynamic Binding RICK NOUWEN ([email protected]) Utrecht Institute of Linguistics (OTS), Utrecht University, NL-3512BL/13 Utrecht, the Netherlands July 3, 2006 Abstract. Quantified appositives have a limited distribution which is reminiscent of quantified discourse anaphora. This article investigates whether this parallel can be fleshed out by means of a two-dimensional dynamic semantics. The proposal works towards an explanation of why quantified appositives are generally infelicitous. Crucial is the fact that variables bound by a strong quantifier have a singular value, while the antecedents for discourse anaphora introduced by the quantifier are plural. Key words: appositives, quantifiers, dynamic binding, multi-dimensional semantics 1. Introduction On many occasions (see del Gobbo 2003b and references therein), a parallel has been observed between discourse anaphoric phenomena on the one hand and constraints on appositives on the other. An illustration of this is presented in (1) and (2). (1) a. b. Jake, a famous Dutch boxer, lives in Utrecht. Jake lives in Utrecht. He is a famous Dutch boxer. (2) a. #Every Dutch boxer, a famous one, took part in the event. b. Every Dutch boxer took part in the event. #He is famous. The example in (1-a) first and foremost informs us that Jake lives in Utrecht. The appositive gives the supplementary comment that Jake is a famous Dutch boxer. The two sentences in (1-b) do something similar. The first one states that Jake lives in Utrecht and the second one adds to this that he is a famous Dutch boxer. While there might be a difference in how we use (1-a) and (1-b), it is clear that what they share is the expression of two propositions, both about properties of Jake. In (2-a), the appositive fails to supply additional information about (the) Dutch boxers. In fact, the appositive does not yield an interpretation at all, and the example is simply infelicitous. In discourse, once again something similar happens. We cannot use the second sentence in (2-b) to express that every Dutch boxer is famous. appositives_dynamic_rolc_versionbeta.tex; 3/07/2006; 14:32; p.1 2 RICK NOUWEN The parallel between discourse and appositive is very consistent. For instance, if we substitute ‘every’ in (2-a) and (2-b) by the indefinite article ‘a’, then both examples are felicitous. (3) a. b. A Dutch boxer, a famous one, took part in the event. A Dutch boxer took part in the event. He is famous. Nevertheless, there is an obvious problem for the comparison between the two kinds of phenomena. In discourse, one can turn to a notion of scope to explain these data. The supplementary nature of appositives severely complicates the notion of scope. In (2-a), for instance, the appositive is sentence internal and, thus, in some sense in the scope of the quantifier. However, the appositive content is invisible to operators in the host sentence. For instance, (4) is not true in a situation where Jake lives in Utrecht but is not a famous Dutch boxer. (4) It is not the case that Jake, a famous Dutch boxer, lives in Utrecht. Still, there is an obvious way in which host sentence and appositive connect. The appositive’s anchor provides a logical subject: in (4), it is Jake who is said to be a famous Dutch boxer. In parenthetical clauses, moreover, the host sentence may provide an antecedent to supplementary material containing an anaphoric pronoun. (5) Jake’s coaches – and, mind you, they worked very hard – were criticised beyond belief. In general, then, apposition and parenthesis is like discourse in that it offers limited access to resources presented in the host/antecedent sentence. However, the fact that appositives are semi-integrated in a sentence makes it difficult to pursue an analysis which does justice to this parallel with discourse. The reason for this is that as long as there is no clearly stated semantics of apposition, we cannot apply our insights from dynamic semantics, the e-type strategy, discourse representation theory or whatever other view we have on the interpretation of discourse anaphora. What is needed is a way of representing the role of appositives and (other) parenthetical material in semantics. Potts 2005 presents a thorough study of what he calls ‘supplements’ by using a multidimensional framework. This means that although the supplementary material is partly integrated in the host sentence, it contributes its content to a separate dimension, thereby escaping the scope of host sentence operators.1 In multidimensional semantics, a sentence S does not result simply in a proposition [[S ]] , but rather in an array of propositions h [[S α ]] , [[S β ]] , . . .i.2 The propositions in the array do not all have the same status. Some of them might be asserted, some might be conventionally implicated, some might be presupposed, some might be separated in a way we do not yet understand. For the purpose of this paper, it does not really matter how these arrays are themselves interpreted. The central topic here is namely to what extent there is an interaction between the dimension appositives_dynamic_rolc_versionbeta.tex; 3/07/2006; 14:32; p.2 APPOSITIVES AND DYNAMIC BINDING 3 associated to an appositive (or a parenthetical clause) and the dimension associated to the host sentence. I will therefore focus on a two-dimensional semantics only and will leave implicit what the (extra) dimensions stands for.3 In this paper, I will apply the idea of two-dimensionality to the dynamic approach to discourse anaphora. In particular, I will spell out the consequences of combining two null assumptions. The first one involves a statement of simplicity: there is no more to the semantics of apposition than two-dimensionality. That is, we expect that if ψ occurs appositively in ϕ, that these two forms are teased apart and that a 2D interpretation results which contains nothing but [[ϕ]] and [[ψ]] . In other words, apposition has no semantic side-effects. The second assumption we start out with is that the distribution of discourse anaphora is correctly accounted for by dynamic semantics. In particular, I assume that a dynamic analysis of quantifiers gives us exactly the anaphoric properties we are interested in. Central to the account of appositives and parentheticals that I propose will be the notion of scope in the semantics. Syntactically, scope is given by scope delimiters such as parentheses. Standardly, parentheses in logic are syntactic instructions that guide interpretation. Normally, there is no meaning assigned to them. Visser and Vermeulen 1996 observe that normally ‘[g]rammar is syncategorematic in [the] approach to semantics’ (p. 322), and that ‘no semantical objects are ascribed to the symbols fixing grammatical structure’ (ibid.). Visser and Vermeulen subsequently aimed at devising formalisms where the role of grammar is categorematical rather than syncategorematical. It goes beyond the scope of this paper to review Visser and Vermeulen’s radical appeal for syntactic flatness in detail. However, I will in part embrace their approach by adhering to the following maxim: do not attribute scope to scopeless actions. By stating a semantics that is fully explicit about scope matters, I will show in detail how appositive content interacts with scope taking operators from the host sentence. The structure of the paper is as follows. Section 2 will present a dynamic semantics for strong generalised quantifiers and will discuss the notion of scope they entail. On the basis of this formalism, in section 3, I construct a simple two-dimensional semantics, which is more expressive only in that it adds a representation of appositive content. Section 4 shows that the dynamic multidimensional formalism readily explains a large part of the data. Finally, in section 5, I discuss some data that is less clearly accounted for. 2. The Dynamics of Quantifiers It is well-known that nominal appositives generally do not occur with quantificational anchors, whereas they are felicitous in combination with referential expressions (McCawley, 1998). (6) a. #Every Dutch boxer, a famous one, takes part in the event. b. #No Dutch boxer, a famous one, takes part in the event. appositives_dynamic_rolc_versionbeta.tex; 3/07/2006; 14:32; p.3 4 RICK NOUWEN c. A Dutch boxer, a famous one, takes part in the event. As I discussed above, the contrast in (6) has a full parallel in the domain of discourse anaphora. Singular pronominal reference to a quantificational antecedent not possible. (7) #Every/A Dutch boxer takes part in the event. He is famous. In dynamic semantics, this contrast between referential and quantificational antecedents is accounted for by assuming a different kind of relational interpretation for these expressions. For existential quantifiers, the so-called donkey equivalence in (8) holds, while for universal quantifiers such an equivalence is invalid. (8) a. b. (9) a. b. ∃x[dutch(x) ∧ boxer(x) ∧ take-part(x)] ∧ famous(x) = ∃x[dutch(x) ∧ boxer(x) ∧ take-part(x) ∧ famous(x)] A Dutch boxer takes part in the event. He is famous. = A Dutch boxer takes part in the event and is famous. ∀x[dutch(x) ∧ boxer(x) ∧ take-part(x)] ∧ famous(x) , ∀x[dutch(x) ∧ boxer(x) ∧ take-part(x) ∧ famous(x)] Every Dutch boxer takes part in the event. He is famous. , Every Dutch boxer takes part in the event and is famous. As is common in dynamic semantics, expressions in the language denote relations between pairs of assignments and worlds. This allows for the simultaneous representation of changes forms make with respect to both knowledge about discourse referents and knowledge of the world. As the discourse progresses less worlds will be accepted and the assignments will contain less valuations to an increasing number of variables. In dynamic predicate logic (DPL, Groenendijk and Stokhof 1991), an existential quantification ∃x[ϕ] represents random assignment to x. It relates assignment functions f to functions that differ from f only in assigning some value to x such that ϕ is verified. It is easy to see that there is really no reason why one should have to define the semantics of existential quantification with respect to some scopal form ϕ. In fact, the donkey equivalence makes clear that a notion of scope is absent with existentials. In the style of dynamic semantics offered by Visser and Vermeulen (Vermeulen, 1993; Visser and Vermeulen, 1996), ∃x is therefore often interpreted as a simple random assignment action: h f, wi [[∃x]] hg, w0 i :⇔ w = w0 & ∀v , x : f (v) = g(v) According to this, ∃x does not tell us anything about the world we are in and it moreover says that any value for x will be accepted. In other words, ∃x resets whatever we knew about x. Subsequent predicates over x may then specify which values for x will be acceptable in which worlds. It is important to realise that there can be no similarly decomposed way in which we may interpret ∀x. Universal quantification only makes sense in combination with the quantifier’s scope. The fact that the donkey equivalence in (9-b) does not hold is a good illustration of this. appositives_dynamic_rolc_versionbeta.tex; 3/07/2006; 14:32; p.4 APPOSITIVES AND DYNAMIC BINDING 5 Consequently, for ∀x we need an interpretation that tests values for x with respect to the whole quantificational structure. h f, wi [[∀x[ϕ][ψ]]] hg, w0 i :⇔ f = g & w = w0 & ∀ f 0 : h f, wi [[∃x ∧ ϕ]] h f 0 , wi → ∃g0 : h f 0 , wi [[ψ]] hg0 , wi A definition like this does justice to the fact that universal quantifiers cannot antecede singular pronouns in discourse. However, it fails to account for the fact that once plural anaphora is taken into consideration everything changes. Examples like (10) show that strong quantifiers are not completely discourse intransparent. (10) Every Dutch boxer will take part in the event. They are all famous. It has been known since Evans 1977 that a non-coreferential, non-bound-variable type of anaphora exists in discourse, where a plural pronoun refers to the maximal set of individuals that satisfy restrictor and scope of quantification. That is, in (10), the plural pronoun refers to (all) the Dutch boxers that will take part in the event. It is fairly straightforward to account for this in dynamic semantics. Given a form ϕ, a world w and an assignment f , let f +w ϕ be the set of output assignments that result from interpreting ϕ with respect to h f, wi. So, f +w ϕ := { f 0 | h f, wi [[ϕ]] h f 0 , wi}. Let G be a set of assignment functions, then tG is the assignment function such that for any variable v: (tG)v = {g(v) | g ∈ G}. In words, t transforms a plurality of assignments into a single assignment to pluralities. Combining these two operations gives us a way of introducing plural entities in the context: t( f +w ϕ) returns a single assignment function which represents all the different output valuations for ϕ in pluralities. For instance, t( f +w (∃x∧ boxer(x))) yields a function that assigns the set of boxers in w to x. A definition like this gives us a way to represent generalised quantification dynamically. Given some contextual f , for a quantified statement Qx[ϕ][ψ] the restrictor set (in w) is given by t( f +w (∃x∧ϕ)), while the intersection between restrictor and scope of the quantifier results from t( f +w (∃x ∧ ϕ ∧ ψ)). On the basis of this, I propose the following dynamic interpretation for the universal quantifier. (Related proposals can be found in work like Krifka 1996, van den Berg 1996, Elworthy 1995, Asher and Wang 2003, Wang et al. 2006, Nouwen 2003, Nouwen 2006.)4 h f, wi [[∀x[ϕ][ψ]]] hg, wi :⇔ w = w0 & g = t( f +w (∃x ∧ ϕ ∧ ψ)) & (t( f +w (∃x ∧ ϕ)))(x) ⊆ g(x) This definition is completely parallel to the standard “Every(A)(B)⇔A⊆ B”. What is more, we may generalise this definition to all strong quantifiers. Assuming that a quantifier symbol Q is interpreted in the model as Q0 , we get: h f, wi [[Qx[ϕ][ψ]]] hg, wi :⇔ w = w0 & g = t( f +w (∃x ∧ ϕ ∧ ψ)) & Q0 (t( f +w (∃x ∧ ϕ))(x) , g(x)) appositives_dynamic_rolc_versionbeta.tex; 3/07/2006; 14:32; p.5 6 RICK NOUWEN Table I. The semantics of LQDPL h f, wi [[ ∃x ]] h f 0 , w0 i :⇔ w0 = w & ∀v , x : f (v) = f 0 (v) h f, wi [[ P(x1 , . . . , xn ) ]] h f 0 , w0 i :⇔ w0 = w & f 0 = f & f, w |= P(x1 , . . . , xn ) h f, wi [[ ϕ ∧ ψ ]] h f 0 , w0 i :⇔ ∃w00 ∃ f 00 : h f, wi [[ ϕ ]] h f 00 , w00 i [[ ψ ]] h f 0 , w0 i h f, wi [[ Qx[ϕ][ψ] ]] h f 0 , w0 i :⇔ w = w0 & f 0 = t( f +w (∃x ∧ ϕ ∧ ψ)) & Q0 (t( f +w (∃x ∧ ϕ))(x) , f 0 (x)) In terms of dynamics, the definition gives an important result. Any occurrence of x in the restrictor or scope of a quantifier Qx is bound by the quantifier. That is, such occurrences are given singular values only. This not only accounts for the fact that co-indexed pronouns in the scope of a quantifier are always bound,5 but also for the fact that strong quantifiers are distributive (see, for instance, Kamp and Reyle 1993 for discussion). Occurrences of x outside the scope of Qx have an exhaustive plural value, which accounts for the famous observation by Evans. To summarise, strong quantifiers differ from existential/referential ones in that they come with a notion of scope. In dynamic semantics, this difference is reflected by the fact that existential quantifiers are actions by themselves, whereas strongly quantified statements have to be interpreted as complexes that include the quantifier’s full scope. As I will show below, the dynamics of quantifiers accounts for a considerable part of the apposition data, given a simple two-dimensional semantics. In fact, the 2D semantics will be nothing more than the quantified dynamic formalism with an added mechanism for dimensionality. That is, the dynamic semantics of quantifiers discussed above gives the bulk of what is needed to account for quantified appositives. I end this section therefore by stating the details of a quantificational version of dynamic predicate logic. The language LQDPL is the language of dynamic predicate logic with generalised quantifiers. The syntax is as follows. First, I assume a set of constants Con. I write Conn for that subset of Con that contains the n-ary constants only. I will furthermore assume that there exists a set of variables Var and a set of terms T erm = Con0 ∪ Var. Finally, there exists a set of (binary) quantifier symbols Quan. ∃v ∈ LQDPL ⇔ v is a variable P(x1 , . . . , xn ) ∈ LQDPL ⇔ P ∈ Conn & x1 , . . . , xn ∈ T erm Qx(ϕ, ψ) ∈ LQDPL ⇔ x ∈ Var & Q ∈ Quan & ϕ , ψ ∈ LQDPL ϕ ∧ ψ ∈ LQDPL ⇔ ϕ , ψ ∈ LQDPL nothing else is in LQDPL Well-formed formulae in LQDPL will be presented on top of a slightly shaded background, this to allow for a clearer distinction between the quantified dynamic logic defined here and the multi-dimensional formalism that will follow. This colour appositives_dynamic_rolc_versionbeta.tex; 3/07/2006; 14:32; p.6 APPOSITIVES AND DYNAMIC BINDING 7 coding will become useful in the semantics of the multi-dimensional logic which, in part, is based on that of LQDPL . The semantics is summarised in table I. 3. 2D Semantics Below, I will introduce a language that will function as a representation for the semantics of appositives. I’ll first provide an informal glance at this language on the basis of (11). (11) a. b. Jake, a famous boxer, lives in Utrecht. ∃x ; x = j ; h ; famous-boxer(x) ; i ; live-in-Utrecht(x) At first sight, this form may seem rather mysterious. But it should be clear that all the relevant information is there. The formula in (10-b) contains both the fact the Jake lives in Utrecht and that he is a famous boxer. The way to read the formal representation in (11-b) is to view it as a series of instructions. That is, I replaced the conjunction symbol with ‘;’ to emphasise the fact that (11-b) is a list of operations as in a complex program. So, ϕ ; ψ corresponds to “first do ϕ, then ψ.” This means that everything apart from these connectives corresponds to an action of some sort. This includes, as I discussed above, ∃x. Moreover, nothing in (11-b) has a scope in the traditional sense of the word. Each of the forms that are connected by semicolons is interpreted by itself as an action on contexts. In what follows, I will use the terms primary and secondary content in order to be able to talk about the different dimensions without obviously adhering to some theoretically assumption on the pragmatic status of the dimensions. The brackets h and i direct the flow of information to the two dimensions. Just like ∃x is a primitive action, so are h and i. The left bracket is an instruction to add future information to the dimension that stores secondary information. The right bracket is an instruction to change back to the dimension of primary content. The action h comes with a presupposition that the current level of interpretation is the standard one, while i presupposes the current level to be the one for apposition. The bracketing in the logical language is therefore not always interpretable. For instance, h; ϕ; h will not receive a defined interpretation in case ϕ does not contain a i, but note that h; ϕ may be defined even if the level is not changed in ϕ. That is, non-matching bracketing is not interpretable, but incomplete bracketing is. Let us now turn to the semantics of the multidimensional language. In dynamic semantics, forms express relations between world-assignment pairs. The simplest way of viewing an example like (11) in terms of relations, is to have the twodimensional language express operations on two completely separated information states. In such an approach, as the discourse unfolds, a doubled context is continuously updated, representing two different ways in which information may be conveyed. Such a view is too simple, however. It leads to a well-known problem with multidimensional formalisms, which was first noticed in Karttunen and Pe- appositives_dynamic_rolc_versionbeta.tex; 3/07/2006; 14:32; p.7 8 RICK NOUWEN ters 1979. For an example like (11), a doubled-context approach would generate two separated updates, one saying that Jake lives in Utrecht, the other saying that there is a famous Dutch boxer. Since the two information dimensions have no access to each other’s referential resources, there is no possibility to do justice to the fact that the appositive’s meaning is referentially related to the meaning of its anchor. The solution is to have dimensions share their referential resources. So, instead of two distinct information states, I take there to be states that consist of one assignment function and two possible worlds. Informally, the interpretation of (11) in some given context, results in triples h f, w, w0 i such that: - f (x) lives in Utrecht in w and f (x) = j - f (x) is a famous boxer in w0 (primary content) (secondary content) In other words, (11) results in triples containing, first of all, a world in which Jake lives in Utrecht and, second, a world in which Jake is a famous boxer. The triple-based semantics for two-dimensionality supports two intuitions: (i) referential resources are shared, and (ii) an appositive expresses a proposition that is independent from the one expressed by the host sentence. As we will see below the sharing of resources has important consequences for the interpretation of quantification into appositives. Before turning to quantification, however, let us see how the current proposal accounts for the second intuition. At each point in discourse, the information that has been conveyed so far is captured in the set of triples that are related to some initial context. Given such a set S , two propositions may be extracted: S 2 = {w | ∃ f, w0 : h f, w, w0 i ∈ S } and S 3 = {w0 | ∃ f, w : h f, w, w0 i ∈ S }. So, for the above example S 2 amounts to Jake lives in Utrecht and S 3 is Jake is a famous boxer. These two propositions represent the information conveyed by host sentences and by supplementary material respectively. However, an agent participating in the discourse might assume that both the proposition expressed by the appositive and the one expressed by the host sentence are true in the actual world. In such cases, the information state needs to collapse. What I mean by this is that when both the appositive and the host are assumed to be true, then there ought to be no difference in the worlds that occur in the two dimensions. There is a straightforward operation that can represent this. Let S be a set of triples h f, w, w0 i: (collapse) !S = {h f, w, w0 i|w = w0 & h f, w, w0 i ∈ S } In what remains, I will ignore the option of information state collapse, however, and assume that for the purpose of studying conditions on appositives one should study the effects of separating out information. One of the merits of the account presented here is that the widest-scope behaviour of appositives follows immediately. For instance, assuming that none of ϕ, ψ and γ contain brackets, then ¬(ϕ; h; ψ; i; γ) will neatly separate the parenthetical meaning ψ (viz. the proposition in S 3 ) from the negative conventional meaning ¬(ϕ; γ) (in S 2 ). (Cf. the discussion of example (4) in the introduction.) appositives_dynamic_rolc_versionbeta.tex; 3/07/2006; 14:32; p.8 APPOSITIVES AND DYNAMIC BINDING 9 I now turn to the details of the 2D semantics. What follows is basically an application of Visser’s logic for polarity switching (Visser, 2002). That is, the existing language of DPL with generalised quantifiers is augmented with control features that allow for the switching between different kinds of content (here, supplementary versus ordinary meaning). Let a possibility be a quadruple hπ, f, ws , wp i, where π ∈ {s, p}, f is an assignment function from variables to (possibly plural) entities and ws , wp possible worlds. The worlds ws will represent the ordinary meaning, while wp represents the meaning that is contributed appositively. The slot π keeps track of what kind of meaning is being processed. Together with the multiplier ×, the meaning types {s, p} form a monoid Π = {{s, p}, ×, s}, where s is the identity symbol and: s×p = p×s s×s = p×p = p = s This monoid is isomorphic to the monoid consisting of 1 and -1 with multiplication (where 1 is the identity symbol and corresponds to s and -1 is the switcher corresponding to p.) In what follows, we will omit the × symbol from our notation and write e.g. ps instead of p × s for p. Let α = hπ, f, ws , wp i be a possibility. We write α p for π, αa for f , αs for ws and αp for wp . The language of LQDPL−2D is that of LQDPL except that we additionally have h and i as atomic formulae. As mentioned before, we will use a background colour to keep the languages apart. The one-dimensional formalism is presented with a shaded background. The first step in interpreting the forms of our multi-dimensional formalism is to provide the link to the quantified dynamic semantics. Let ϕ be an atomic form in LQDPL . q[{ϕ}]q0 :⇔ q0p = q p & q0pq p = qpq p & hqa , qq p i [[ ϕ ]] hq0a , q0q p i What this says is that ϕ is interpreted in LQDPL−2D almost exactly as it is in LQDPL . The only difference is that it depends on the value of q p whether the information conveyed by ϕ ends up in the s or in the p-slot. If q p = s than the q0s slot might differ from that of qs , whereas q0p will remain the same as qp . If, on the other hand, q p = p than the p-information will change, while the s part will be left untouched. For the extension LQDPL−2D offers to LQDPL , we will need the following additional semantics. q {[h}] q0 :⇔ q0p = p & q0a = qa & q0s = qs & q0p = qp q {[i}] q0 :⇔ q0p = s & q0a = qa & q0s = qs & q0p = qp q[{ϕ; ψ}]q0 :⇔ ∃q00 : q[{ϕ}]q00 {[ψ}]q0 The interpretation of the brackets is straightforward. They change the value of the slot that keeps track of what kind of information is being processed. Brackets come appositives_dynamic_rolc_versionbeta.tex; 3/07/2006; 14:32; p.9 10 RICK NOUWEN with the presupposition that they actually do change something. So, h cannot be performed in a context where the p is already the current mode. Similarly, i is unsuitable in context q should q p be s. So, with ↓ expressing definedness: q {[h}] q0 =↓ :⇔ q p = s q {[i}] q0 =↓ :⇔ q p = p q[{ϕ; ψ}]q0 =↓ :⇔ ∃q00 : q[{ϕ}]q00 =↓ & q00 {[ψ}]q0 =↓ Here are some simple examples. Assume ϕ , ψ etc. are well-formed formulae from the language LQDPL (so they do not contain brackets). Assume moreover that ϕ and ψ do not contain quantifiers over the same variable. Let q be an arbitrary information state with q p = s: q = hs, f, ws , wp i. The following holds (I use ↑ to express undefinedness): (12) q[{ϕ; h; ψ}]q0 ⇔ q0p = p & h f, ws i [[ ϕ ]] hq0a , q0s i & h f, wp i [[ ψ ]] hq0a , q0p i (13) q[{ϕ; h; ψ; h}]q0 = ↑ for every q and q0 (14) q[{ϕ; h; ψ; i; ϕ0 ]}q0 ⇔ q0p = p & h f, ws i [[ ϕ; ϕ0 ]] hq0a , q0s i & h f, wp i [[ ψ ]] hq0a , q0p i Since quantified sentences are interpreted as syncategoremata, we need to restate their semantics in the extended language. Before we can do this, however, we first of all need to redefine the ‘+0 -operation introduced above for the summation of assignment functions, so that it can deal with multidimensional information states. We write +q ϕ for the set of assignment functions that result in an information state after interpreting ϕ w.r.t. q. Formally: +q ϕ := {q0a | q[{ϕ}]q0 } I furthermore define an operation which tests whether the content of a nonactual dimension is compatible with a possibility. This will be useful to test whether the appositive content (if any) within the scope of a quantifier is compatible with an output possibility created by quantification. ∃w : hq p , qa , w, qp i[{ϕ}]hq0p , q0a , w, q0p i & qp = q0p ! 0 q[{ϕ}] q :⇔ ∃w : hq , q , q , wi[{ϕ}]hq0 , q0 , q0 , wi & q = q0 p a s p a s s s if q = s if q = p Given a possibility q with q p = s, an action {[(∃x; ψ; h; ϕ; i; ψ0 )}]! is successful only if the secondary world (qp ) is such that the value for x provided by qa is compatible with ϕ. Moreover, any variable-introductions embedded in ϕ end up in the output assignment. With respect to the primary content dimension, the definition merely states the condition that there exists a world for which the relation in question holds, which is a triviality. appositives_dynamic_rolc_versionbeta.tex; 3/07/2006; 14:32; p.10 11 APPOSITIVES AND DYNAMIC BINDING The semantics of a quantificational sentence is now as follows. q[{Qx[ϕ][ψ]}]q0 :⇔ q p = q0p & qs = q0s & qp = q0p & ∃ f, g : f = t(+q (∃x; ϕ; ψ)) & g = t(+q (∃x; ϕ)) & Q0 (g(x) , f (x)) & hq p , f, qs , qp i[{ϕ; ψ}]! q0 This is completely parallel to the dynamic semantics of generalised quantifiers given above. The only difference is that it incorporates forms from the 2D language instead of forms from the original dynamic logic and that it tests whether its output is compatible with any content for the non-actual dimension which might be embedded in the quantifier’s scope.6 Informally, this definition yields the following separation of conditions: - The set of atomic values for x that would yield an output for interpreting ϕ is in the Q0 relation to the set of atomic values for x that would yield an output for interpreting ϕ; ψ. The value for x is this latter set. (primary content) - Any appositive conditions that exist within the form ϕ; ψ are compatible with the output assignment and qp . (secondary content) The assignment function resulting from interpreting a quantificational statement is the result from combining possible atomic extensions of the input assignment. However, such extensions might involve appositive conditions on the value of the variable. Since the two dimensions share their referential resources, the quantifier counts value assignments not only constrained by the actual, but also by the other dimension. So, importantly, the resulting value for x is restricted by conditions from all dimensions. As we will see below, this has important consequences for the distribution of quantified appositives. 4. Multidimensionality and Quantified Appositives I start with the case of existential anchors.7 (15) a. b. A Dutch boxer, a famous one, won the tournament. ∃x; dutchboxer(x); h; famous(x); dutchboxer(x); i; won(x) The relation expressed by the form in (15-b) contains pairs of quadruples hq, q0 i where: q p = q0p = s and where q0a is such that it assigns to x a value that is a Dutch boxer who won the tournament in the world q0s and who is a famous Dutch boxer in world q0p . In less formal terms: - the value for x is a Dutch boxer who won - the value for x is a famous Dutch boxer (primary content) (secondary content) appositives_dynamic_rolc_versionbeta.tex; 3/07/2006; 14:32; p.11 12 RICK NOUWEN Note that (15-b) already yields an important result. Existentially anchored appositives are interpretable and result in separated conditions on the same variable assignment. Turning now to a quantificational variation on (15-a):8 (16) a. #Every Dutch boxer, a famous one, took part in the tournament. b. Every x[dutchboxer(x)] [h; famous(x); dutchboxer(x); i; tookpart(x)] Informally, this results in the following: - the value for x is the set of (famous) Dutch boxers who took part in the tournament; this set has the set of Dutch boxers as a subset (primary content) - the value for x is a famous boxer (secondary content) There can be no such value assignment, since one dimension demands the value to be a plurality, whereas the other demands it to be singular. Consequently, the form in (16-b) denotes the empty relation. Strong quantifiers compare possible valuations for variables. So, (16-b) yields an output possibility if and only if all values d for x, such that d is a Dutch boxer in the s-world are such that d is famous in the p-world and took part in the tournament in the s-world. Because of this comparison of valuations for the restrictor and scope, any supplementary information will be taken into account as well. In other words, appositive occurrences of x in the scope of a strong quantifier Qx are bound. The quantifier summates the singular values for its running variable into one maximal antecedent for future anaphora. This is in conflict, however, with the fact that the appositive content is supposed to hold independent of the quantificational relation. In other words, the plural value for x that results from quantification is incompatible with the appositive condition that x be a single famous boxer. The above suggests that quantified appositives are only infelicitous because the quantifier wants to set up (plural) exhaustive reference for the quantified variable. This implies that when the appositive’s content predicates over a variable other than the quantifier’s running variable, the apposition can be quantified without problems. In natural language, this prediction is borne out by examples like (17), first observed by Wang et al. (2005). (17) If a professor, a famous one, writes a book, he will make a lot of money. The most salient reading for (17) is one that says that famous book-writing professors make a lot of money. In discourse, a conditional like that in (17) does not give rise to exhaustive reference to the set of professors that wrote a book and made a lot of money: (18) If a professor writes a book, he will make a lot of money. #They are famous. appositives_dynamic_rolc_versionbeta.tex; 3/07/2006; 14:32; p.12 APPOSITIVES AND DYNAMIC BINDING 13 Such a conditional, represented here as (ϕ) ⇒ (ψ), ought to receive a simple static interpretation that tests whether all cases in which ϕ holds are cases in which ψ hold. For (17), we arrive at: (19) (∃x; professor(x); h; famous(x); professor(x); i; ∃y; book(y); write(x, y)) ⇒ (makealotofmoney(x)) The appositive says that values for x should be famous professors. The conditional says that all cases in which a professor x writes a book, are cases in which this x makes a lot of money. The two dimensions together result in a further restricted reading which says that famous book-writing professors make a lot of money. In sum, since strong quantifiers involve the comparison of possible value assignments that result from their scope, any apposition included in this scope domain will influence the comparison. The fact that outside the quantifier’s scope appositive and quantifier pose conflicting conditions on the variable explains why appositives are generally not quantified. As we will see next, however, the data are not as straightforward as this. 5. Discussion of further data Recently, it has been observed that, in some cases, appositives come with quantificational anchors. Potts 2005 (p. 124), for instance, gives the following example. (20) Every climber, all experienced adventurers, made it to the summit. There is a problem, however, with the intuitions behind an example like this. With universal quantification it is impossible to distinguish two kinds of anaphoric options that result from quantifiers: one being exhaustive reference to the intersection of the quantifier’s restriction and scope, the other being reference to the presupposed domain of quantification. Consider, for instance, (21-a) and (21-b). (21) a. b. c. Most students attended the conference. Some of them presented a paper. Most students failed to show up at the conference. Some of them, however, came and presented a paper. Most semanticists in the Dutch department praised their joint research proposal. In (21-a) the most natural way to understand the pronoun ‘them’ is for it to refer to the students that attended the conference. In (21-b), however, another kind of antecedent is picked up by the pronoun. In the second sentence of (21-b), the pronoun cannot refer to the intersection of restrictor and scope of the antecedent sentence, for this would result in the contradictory claim that some of the students that failed to show up at the conference, did show up and presented a paper. The pronoun instead refers to the domain of quantification of the antecedent sentence, namely appositives_dynamic_rolc_versionbeta.tex; 3/07/2006; 14:32; p.13 14 RICK NOUWEN the set of (salient) students. This type of reference is sometimes called reference to the maximal set, or, in short, maxset reference. A similar reading is available for (21-c). As we can observe from (21-b) and (21-c), this type of anaphora can occur both within and outside the scope of quantification. Presumably, this is so because strong quantifiers presuppose their domain. For universal quantifiers like that in (20), however, there is no difference between referring to the maxset (the set of climbers) or the set of climbers that made it to the summit, since the sentence itself claims these two sets are identical. And so, we could claim that (20) is a case of maxset reference, which given the fact that this type of reference is felicitous within the scope of the quantifier, renders (20) completely unproblematic. It is quite easy, however, to come up with examples involving non-universal quantifiers. For instance, the appositive in (22) is quantified without this leading to the example being infelicitous. (22) Less than half the climbers, all French nationals, made it to the summit. On at least one reading, this example says that less than half the climbers made it to the summit and that all these successful climbers were French nationals. This reading is comparable to a discourse anaphoric reading where a pronoun refers to the intersection of restriction and scope, exactly the kind of anaphora I have claimed did not exist within the scope of a quantifier. The problem is worse still, since given the semantics for apposition we gave above, we would expect that if felicitous at all, an example like (22) would result in an interpretation where the appositive actively participates in the comparison between sets denoted by the quantifier. In other words, we would expect, (22) to say that less than half the climbers were French national who made it to the summit. Of course, this is not what the example conveys. A clue to the solution of this problem is the fact that the appositive in this case is plural. This means it could never be bound by the quantifier, since the quantifier only takes singular values into account. (Recall that strong quantifiers are distributive. Compare also to footnote 10.) This suggests that, initially, the appositive is not co-indexed with the quantifier, and so ends up not being bound at all. (23) < 21 x[climber(x)][h; ∃y; french-pl(y); i; reachsummit(x)] Reference resolution may then at a later stage connect the plurality formed by the quantifier with the plurality described by the appositive. Only when this happens after the quantificational sentence has been processed is this possible. So, (24-a) results in an empty relation, because there is no way the plural value for y can be identical to the singular value for x, but (24-b) gives us exactly the reading we are after. (Here, ’french-pl’ is an abbreviation of a test on y that is successful only if y is a plurality of French nationals). (24) a. < 12 x[climber(x)][h; ∃y; french-pl(y); x = y; i; reachsummit(x)] appositives_dynamic_rolc_versionbeta.tex; 3/07/2006; 14:32; p.14 APPOSITIVES AND DYNAMIC BINDING b. 15 < 12 x[climber(x)][h; ∃y; french-pl(y); i; reachsummit(x)]; x = y Note that a similar strategy is not possible for singular appositives. The strategy corresponding to (24-b) would in that case be out because of the clash between singular y and plural x. The strategy corresponding to (24-a) would be equally infelicitous because the appositive content contains the condition that the values of x and y are identical. This identity cannot be maintained outside the scope of the quantifier, however. An approach like the one in (23) only works for nominal appositives, however. As soon as we take a close look at appositive relatives, it becomes clear that not all appositions are alike. (25) ??Most students, who (all) arrived very late, brought a bottle of wine. Del Gobbo, however, observes that the position of an appositive relative matters. (26) a. #Many students, who arrived very late, brought a bottle of wine. b. They invited many students, who arrived very late. (del Gobbo 2003a, p. 26) According to (26-b), many students were invited and all of the invited students arrived very late. An obvious problem with the contrast in (26), however, is that it is not clear at all that many is a strong quantifier. It is well-known that one of the uses of many is weak, as in There are many students here. This means that one could argue that (26-b) happens to trigger the weak reading, whereas the subject position of (26-a) results in a strong reading for many. But if we use an unambiguously strong quantifier in many’s place, then it becomes apparent that the contrast in (26) is a more general phenomenon. (27) a. b. Less than half the climbers, who (by the way) were (all) French nationals, made it to the summit. They interviewed less than half the climbers, who (by the way) were (all) French nationals. It seems that for (27-a), only a maxset reading is available (that is, saying that all climbers are French), whereas for (27-b) a reading exists where the interviewed climbers are all French. If this data is trust-worthy, then this may show that the referential link between an appositive relative and its anchor is stronger than the link between a nominal appositive and its anchor. What I mean is that relatives seem to have to be co-indexed with their anchor.9 The 2D semantics developed above then suggests a straightforward account of the contrast in (27), based on the assumption that sentence-final appositives can be interpreted outside the quantificational structure. So, (27-a) and (27-b) correspond respectively to (28-a) and (28-c). (28) a. < 12 x[climber(x)][h; french-pl(x); i; reachsummit(x)] appositives_dynamic_rolc_versionbeta.tex; 3/07/2006; 14:32; p.15 16 RICK NOUWEN b. c. < 12 x[climber(x)][reachsummit(x); h; french-pl(x); i] < 12 x[climber(x)][reachsummit(x)]; h; french-pl(x); i The form in (28-b) is a possible logical form for (27-b), but it is rejected because it results in denoting the empty relation, just like (28-a). The form that corresponds to (27-a) in its felicitous reading is (28-a) but with french-pl(x) replaced by predication over whatever variable corresponds to the presupposed set of climbers. 6. Conclusion The formalism presented in this paper facilitates a comparison between discourse anaphora and quantification into appositives. The result is that if we stick to a semantics that yields the proper predictions for anaphoric possibilities in discourse, the most unshakeable observations on quantified apposition can be explained in terms of multidimensionality and the sharing of referential resources. However, as the last section shows, the landscape of referential possibilities is never easily captured. I showed that the proposal allows for amendments that yield desirable results. Undoubtedly, however, there will be extra-semantic aspects of apposition that will not be accommodated so easily. Syntactic, discoursal and pragmatic constraints, and their interaction with semantics, have been largely overlooked in this paper and have to be left to further research.10 Acknowledgements Earlier versions of this paper have been presented at the DGfS workshop on Parentheticals in Bielefeld in February 2006 and the LENS workshop in Tokyo in June 2006. I am indebted to the audience of these workshops for valuable feedback. I would like to thank Carla Umbach, Alastair Butler, Eric McCready, Linton Wang, Christopher Potts and an anonymous reviewer for helpful comments and thought-provoking critical remarks, some of which I was unfortunately not able to address in this article. Any mistakes are mine. This work was funded by a grant from the Netherlands Organisation for Scientific Research (NWO), which I hereby gratefully acknowledge. Notes 1 The use of separate dimensions in semantics is not new. The locus classicus is Karttunen and Peters 1979; other examples include the focus-semantics literature based on Rooth 1985 and the many-dimensional representation framework proposed by Geurts and Maier (2003). 2 For the sake of exposition, I focus on the propositional level here only. In a compositional derivation, however, multidimensionality applies to any model-theoretic type. 3 One possible characterisation of the added dimension is to see it as a place where conventional implicatures are stored. The link between appositives and conventional implicatures (Grice, 1975) is discussed in detail by Potts (2005) (see also Jayez and Rossari 2005). appositives_dynamic_rolc_versionbeta.tex; 3/07/2006; 14:32; p.16 APPOSITIVES AND DYNAMIC BINDING 17 4 Note that this definition predicts that any existential quantifier that may occur in the restrictor ϕ or scope ψ yields maximal anaphora in discourse. This prediction is borne out by examples like: “Every student in the Dutch class wrote a paper. {#It wasn’t/They weren’t} well-written, though.” 5 This can be demonstrated on the basis of the following examples from Kamp and Reyle 1993. (a) Two lawyers (each) hired a secretary they interviewed. (b) Most lawyers hired a secretary they interviewed. The example in (a) is ambiguous. In the first reading, both lawyers hired a secretary he or she interviewed. In the second, they each hire a secretary that was interviewed by both of them. This ambiguity involves the possibility of the plural pronoun ‘they’ to be construed as a collective subject for ‘interview’. In other words, in the first reading the secretary was interviewed by the lawyer that hired him, while in the second he was interview by the two lawyers. In contrast to (a), (b) lacks this latter reading. One cannot use (b) to convey that a majority of lawyers each hired a (different) secretary they collectively interviewed. 6 This final condition might seem a superfluous at first sight, but it is not. Without it the dynamics of the non-actual dimension will be completely lost. The test whether the quantificational relation holds is fully based on the summation of successful assignment functions. But, if we happen to evaluate this relation with respect to a possibility in which the appositive content within the quantifier’s scope is false, this could potentially verify the quantificational statement. For instance, an example like “No friend of Sue, who by the way is Cody’s sister, is rich” should not turn out true in a possibility in which all of Sue’s friends are rich, but Sue is not Cody’s sister. The final condition in the semantics for quantified sentences eliminates such possibilities. 7 For ease of exposition, I assume that the ‘one’-anaphor can be interpreted as a repetition of the noun phrase ‘Dutch boxer’. 8 In (29), I represent the appositive as part of the nuclear scope of the quantifier. It is not obvious at all that this is correct. However, for the analysis that follows this choice is immaterial. What matters is that both the appositive and the host sentence give conditions on the same variable, which eventually leads to a clash. 9 Assuming that the anchor carries an index for both the maximal set (the domain of quantification) and the running variable of quantification, this predicts that relatives are either bound or domainreferring. 10 One of the most pressing issues appears to be the interaction of apposition with presupposition. Eric McCready and Linton Wang (p.c.) provided me with the following example: “John, a good tennis player, is a good golfer too”. Admittedly, it is not immediately clear how such examples are to be treated in the dynamic formalism provided here. Much will depend, however, on the more general question of what exactly the pragmatic status of appositive content is, which is an issue I have not commented on at all in this paper. References Asher, N. and L. Wang: 2003, ‘Ambiguity and Anaphora with Plurals in Discourse’. In proceedings of: Semantics and Linguistic Theory 13 (SALT 13), University of Washington, Seattle, Washington. van den Berg, M.: 1996, ‘Some aspects of the internal structure of discourse: the dynamics of nominal anaphora’. Ph.D. thesis, ILLC, Universiteit van Amsterdam. Elworthy, D.: 1995, ‘A theory of anaphoric information’. Linguistics and Philosophy 18, 297–332. Evans, G.: 1977, ‘Pronouns, Quantifiers and Relative Clauses’. Canadian journal of philosophy 7, 467–536. Geurts, B. and E. Maier: 2003, ‘Layered DRT’. Unpublished Manuscript, Radboud University Nijmegen. appositives_dynamic_rolc_versionbeta.tex; 3/07/2006; 14:32; p.17 18 RICK NOUWEN del Gobbo, F.: 2003a, ‘Appositives and Quantification’. In: E. Kaiser and S. Arunachalam (eds.): Proceedings of the 26th Annual Penn Linguistics Colloquium, Vol. 9.1 of Pennsylvania Working Papers in Linguistics. del Gobbo, F.: 2003b, ‘Appositives at the Interface’. Ph.D. thesis, University of California, Irvine. Grice, P.: 1975, ‘Logic and Conversation’. In: P. Cole and J. Morgan (eds.): Syntax and Semantics, 3: Speech Acts. New York: AP. Groenendijk, J. and M. Stokhof: 1991, ‘Dynamic Predicate Logic’. Linguistics and Philosophy 14, 39–100. Jayez, J. and C. Rossari: 2005, ‘Parentheticals as conventional implicatures’. In: F. Corblin and H. de Swart (eds.): Handbook of French Semantics. Stanford: CSLI Publications. Kamp, H. and U. Reyle: 1993, From Discourse to Logic. Dordrecht: D. Reidel. Karttunen, L. and S. Peters: 1979, ‘Conventional Implicature’. In: C.-K. Oh and D. A. Dinneen (eds.): Presupposition, Vol. 11 of Syntax and Semantics. New York: Academic Press. Krifka, M.: 1996, ‘Parametrized Sum individuals for plural reference and partitive quantification’. Linguistics and Philosophy 19, 555–598. McCawley, J.: 1998, The Syntaxtic phenomena of English. Chicago: Chicago University Press. Nouwen, R.: 2003, Plural pronominal anaphora in context, No. 84 in Netherlands Graduate School of Linguistics Dissertations. Utrecht: LOT. Nouwen, R.: 2006, ‘On Dependent Pronouns and Dynamic Semantics’. Journal of Philosophical Logic. to appear. Potts, C.: 2005, The Logic of Conventional Implicatures, Vol. 7 of Oxford Studies in Theoretical Linguistics. Oxford University Press. Rooth, M.: 1985, ‘Association with Focus’. Ph.D. thesis, University of Massachusetss at Amherst. Vermeulen, K.: 1993, ‘Sequence semantics for dynamic predicate logic’. Journal of Logic, Language and Information 2, 217–254. Visser, A.: 2002, ‘The Donkey and the Monoid. Dynamic Semantics with Control Elements’. Journal of Logic Language and Information 11(1), 107–131. Visser, A. and C. Vermeulen: 1996, ‘Dynamic bracketing and discourse representation’. Notre Dame Journal of Formal Logic 37, 321–365. Wang, L., E. McCready, and N. Asher: 2006, ‘Information Dependency in Quantificational Subordination’. In: K. von Heusinger and K. Turner (eds.): Where Semantics meets Pragmatics, Crispi 16. Elsevier. Wang, L., B. Reese, and E. McCready: 2005, ‘The projection problem of nominal appositives’. Snippets 10, 13–14. appositives_dynamic_rolc_versionbeta.tex; 3/07/2006; 14:32; p.18
© Copyright 2026 Paperzz