Non-local attachment of clauses

Abstract

We argue that some parenthetical-like clauses in ASL can take both intermediate and maximally wide scope outside of if-clauses and attitude verbs. Specifically, we investigate embedded coordinations, of the form … SAY [IF Clause-1 Clause-2 PLUS Clause-3, …], and argue that Clause-2 may in some cases be interpreted with wide and intermediate scope (above SAY, or between SAY and IF). The key to our paradigm is that we mark the scope of IF and SAY with non-manual markers (Brow Raise and/or Role Shift). By exempting Clause-2 from these non-manuals, we force it to outscope the relevant operator, including when it might be expected to create a syntactic island. Wide scope replicates the behavior of parentheticals and appositives in other languages. Intermediate scope is particularly interesting because it mirrors with full clauses the behavior predicted by some theories for some English appositives (Schlenker 2010, to appear a). The ASL data might thus lend support to the existence of a mechanism of high attachment in situ (McCawley 1981, 1998). Alternatives in terms of island-escaping covert movement or in situ indexing face significant challenges.

Appendix: Problems with a movement-based account

We explore in greater detail the possibility that wide and intermediate scope readings of the ASL clauses studied in this piece arise through covert movement. This analysis faces two challenges: it might require movement out of some syntactic islands; and it might not yield the right meanings.

□:

Syntactic islands

A movement account of wide and intermediate scope readings is initially implausible because the movement would need to cross a classic island, namely an IF-clause. But as we now show, the situation is more complicated than it seems, because our consultant tends to accept wh-extraction out of the relevant IF-clauses. On the other hand, extraction out of coordinate structures is sharply degraded. Thus if we take the extraction in (13)b and possibly (13)c (repeated as (41)b,c) to be out of a coordinate structure island in addition to an IF-clause, a movement analysis will prove implausible.

• (41)

  

In (42)b and (43)b, wh-extraction out of an IF-clause is only minimally degraded relative to the controls, including ones with a resumptive pronoun, as in (42d)–(43d). The constructions we picked are not agreement verbs (also called ‘indicating verbs’ or ‘directional verbs’), which are independently known to obviate some binding-theoretic effects (Lillo-Martin 1991). But it is worth noting that Koulidobrova (2017) has further examples of apparently licit extraction out of an island in the absence of agreement or resumption.

• (42)

  

• (43)

  

By contrast, extraction out of a coordinate structure with an overt AND is degraded for our consultant, as seen in (44)a,b and (45a,b. These contrast with the acceptable versions involving resumptive pronouns in (44)c,d–(45)c,d.²⁷

• (44)

  

• (45)

  

The same results are obtained with asyndetic coordinate structures as in (46)b,c and (47)b,c (we do not know why (46)b is slightly more acceptable than (47)b).

• (46)

  

• (47)

  

In sum, no clear island effects are obtained with wh-extraction out of IF-clauses alone, but clear violations are obtained for wh-extraction out of coordinate structures, with or without an overt AND. Thus if we take the extraction in (41)b,c to be out of a coordinate structure island, we would expect it to be degraded, contrary to the judgments our consultant obtained. This is unexpected for an analysis based on covert movement, unless covert movement can somehow escape the coordinate structure island constraint.

□:

Meaning

A movement analysis faces another challenge: it is not clear that it yields the right meaning. Consider for instance the wide scope reading obtained in (26)c, repeated as (48):

• (48)

  

This sentence is schematically represented in (49)a. Let us take Clause-2 to be (as is natural) of propositional type, i.e. <s, t>, where s is the type of worlds and t is the type of truth values. On the standard assumption that movement leaves behind a trace bound by a λ-abstractor (Heim and Kratzer 1998), the resulting structure could be interpreted as in (49)b, where the trace is of intensional type, i.e. <s, t>, or conceivably as in (49)c, where the trace just denotes a truth value, and is thus of type t (we assume that parataxis is interpreted as conjunction).

• (49)

  

Let us consider each possibility in turn, starting with (49)b. The problem is that in this case the movement won’t affect interpretation at all. To make the technical point concrete, let’s consider a simpler case, with the semantics of conditionals in (50) within a modal semantics with a world parameter w (and an assignment function s).

• (50) 〚if F, Q〛^{w, s} = 1 iff for every world w′ accessible from w, if 〚F〛^{w′, s} = 1, 〚Q〛^{w′, s} = 1.

In this simple example, the possibility we are considering is that F is moved out if the if-clause and abstracted over as an object of type <s, t>, as in (51):

• (51) F λp_{<s, t>} [if p, Q]

We need a rule to interpret this structure, and the following version of Intensional Function Application (e.g. Heim and Kratzer 1998) will do:

• (52) Intensional Function Application (special case) If 〚F〛^s is of type <s, t> and 〚G〛^{w, s} is of type <<s, t>, t>, 〚F G〛^{w, s} = 〚G〛^{w, s}(〚F 〛^s).

The problem is that the fact that F appears outside the if-clause does not change the truth conditions, because in (52) the value of F is not sensitive to the value of the world parameter (as is standard, we write s[p→π] for the assignment function which is identical to s except that it assigns π to the variable p):

• (53)

  	=	(by applying (52))
  	= [λπ<s, t> 〚if p, Q〛w, s[p→π]](〚F 〛s)
  	= 〚if p, Q〛w, s[p→π] with π= 〚F 〛s
  	= 〚if F, Q〛w, s

The same problem arises in (49)b as in (51), with the difference that the sentence is far more complicated, as it involves extraction out of a conjunction which is itself embedded within an if-clause under say. But the difficulty is the same: it is unclear how the movement of Clause-2 can affect the truth conditions.

The second possibility is to take the abstraction to be over an expression of type t rather than <s, t>, as in (49)c. This means that we must use a version of Extensional Function Application (e.g. Heim and Kratzer 1998), as in (54).

• (54) Extensional Function Application (special case)

  If 〚F〛^{w, s} is of type t and 〚G〛^{w, s} is of type <t, t>, 〚F G〛^{w, s} = 〚G 〛^{w, s}(〚F 〛^{w, s}).

To see the effect of the rule in a much simpler case, we consider the extraction in (55), where the propositional variable p now has type t rather than <s, t> (as in (51).

• (55) F λp_t [if p, Q]

Contrary to the case in (53), moving F will now have an effect:

• (56)

  $[\!\![\text{F $\lambda $p$_{\mathrm{t}}$ [if p, Q]}]\!\!]$w, s	= $[\!\![\text{$\lambda $p$_{\mathrm{t}}$ [if p, Q]}]\!\!]$w, s(〚F 〛w, s)	(by (54))
  	= [$\lambda \pi _{\mathrm{t}}$ 〚if p, Q〛w, s[p→π]](〚F 〛w, s)
  	= 〚if p, Q〛w, s[p→π] with π = 〚F 〛w, s

Now the truth value of F at w will be rigidly assigned to the variable p. Assuming for the sake of argument that we can proceed with the computation,²⁸ we will assign a specific value to π = 〚F 〛^{w, s}. For concreteness, we take it to have the value false, i.e. 0. The computation then proceeds as in (57):

• (57) With π = 〚F 〛^{w, s} = 0, 〚if p, Q〛^{w, s[p→π]} = 1

  iff for every world w′ accessible from w, if 〚p〛^{w′, s[p→π]} = 1, 〚Q〛^{w′, s} = 1,

  iff for every world w′ accessible from w, if s[p→π](p) = 1, 〚Q〛^{w′, s} = 1,

  iff for every world w′ accessible from w, if π = 1, 〚Q〛^{w′, s} = 1,

  iff for every world w′ accessible from w, if 0 = 1, 〚Q〛^{w′, s} = 1.

The last line is trivially satisfied, and this is a bad result: the fact that F is false in the world of evaluation makes the conditional with the moved antecedent trivially true. This problem arises in a more complex setting in (49)a analyzed as in (49)c. On the intended reading, (49)a ought to entail Clause-2, hence (49)a couldn’t be true while Clause-2 is false. But with the analysis we just sketched, this is not the case; the falsity of Clause-2 in the actual world will wrongly lead to the prediction that Ann said something trivially true (informally, the predicted meaning for (49)c is akin to: Ann says that if Clause-1 and falsity and Clause-3, Consequent).

To get out of this problem, one would need to posit, with various analyses (e.g. Schlenker 2004, among others), that a conditional whose antecedent is trivially false yields a presupposition failure. The presupposition failure of the conditional could then be expected to percolate to the matrix sentence (Heim 1992). Thus from the assumption that the asserted sentence is true, one would get the inference that Clause-2 is true in its world of evaluation—on a wide scope reading, the actual world. The full repercussions of this analysis would need to be investigated, however.

A third alternative is to posit that Clause-2 moves out without leaving behind an interpreted trace, as schematized in (49d) (with the continued assumption that parataxis is interpreted as conjunction). This might conceivably explain why this movement does not display the hallmarks of the coordinate structure constraint, since no trace is left within the syntactic island. But it remains mysterious how this operation fits with standard views on the syntax/semantics interface for movement structures.

Finally, we note that the same problems and solutions are relevant for the intermediate scope reading of Clause-2, as illustrated in (58). On the intended reading, what Ann says entails Clause-2. But on bivalent analyses of conditionals, (58)a may be true on the analysis in (58)c even though in each world compatible with what Ann says, Clause-2 is false, as this makes the embedded conditional trivially true. The presuppositional solution discussed above could presumably be applied to this case as well.²⁹

• (58)

  

In sum, a covert movement analysis of intermediate (and wide) scope readings has two problems to address: the movement might have to be out of a coordinate island; and it might or might not yield the right inferences depending on the details of one’s analysis.