SemPrag03.Progr.pdf - Institut für Linguistik/Germanistik - Universität ...

meaningful way to connect the two sentences. In SDRT terms, this translates as an inability to infer
any discourse relation between the two segments. We spell this idea out formally for the similar case
(1), the various segments of which are translated as follows in the formalism presented in section 2:
(1 ′ ) a. ∃x; all x (hunter(x)); j Dep (x, y); D(((∃y; deer(y); saw(x, y)) → shot(x, y)))
b. singular(y); died immediately(y)
c. singular(y); a y (female(x))
The theory of SDRT update (which includes a theory of discourse attachment, of which space
considerations preclude a detailed discussion) gives the following meaning to (1a,b). Here Result is
a rhetorical relation which requires a causal relation between two bits of information:
(11) ∃x; all x (hunter(x)); j Dep (x, y); D(((∃y; deer(y); saw(x, y)) → Result(shot(x, y), singular(y);
died immediately(y)))
The above may be interpreted roughly as For every hunter 1 , in case that that hunter 1 saw a deer 2 ,
he 1 shot it 2 and then caused it 2 to die immediately, which is intuitively correct. For (1a,c), however,
no rhetorical relation can be inferred since the necessary background for this inference is lacking
(with additional context, however, such an inference may be forthcoming, improving the discourse).
Thus, the following coherent logical form cannot be constructed for any relation R for (1a,c):
(12) ∗ ∃x; all x (hunter(x)); j Dep (x, y); D(((∃y; deer(y); saw(x, y)) → R(shot(x, y), singular(y); a y (female(y)))
Similar facts hold for (10), explaining its infelicity.
Thus, on our account, the infelicity of the impossible cases of telescoping follow directly from
independently necessary constraints on the coherence of discourse, not from any ad hoc constraints
that must be defined on the availability of special update mechanisms. This account is related to,
but not identical with, the script-based account of Poesio & Zucchi (1992); unlike them, however, our
analysis is not specific to quantificational subordination, but falls out of general pragmatic constraints
on possible discourses. Our analysis also incorporates a distinction between the information released
by various universal quantifiers, which is overlooked in their analysis.
4. Discourse Subordination Re-Generalized
The account presented here can be generalized to other kinds of discourse subordination. In
quantificational subordination, licensing depends on compatibility between the object introduced by a
universal quantifier and the pronoun which depends on it. Abstracting away from the quantificational
case, subordination comes with a compatibility requirement between a dependent object and its
antecedent. In the modal domain, this requirement corresponds to a need for the domains of modal
operators to ‘fit’ one another. In the literature, it has been noted that certain types of modality
work together, and others do not:
(13) a. A wolf might come in. It would eat you first. (epistemic; subjunctive)
b. John should buy a car. He would drive it (deontic; subjunctive)
(14) a. A wolf will come in. # It would eat you first. (indicative; subjunctive)
b. John might buy a bottle of wine. # He should drink it. (epistemic; deontic)
In our terms, the felicity/infelicity of the above examples corresponds to the compatibility of the
modal operators involved.
This observation itself is not new; scholars of modal subordination (Roberts, Frank, Geurts) have
all noted it in one form or another. However, we think that our theory provides a new perspective
on the facts by claiming that this need for compatibility is not limited to the modal case, but rather
holds for all sorts of subordinating contexts. We also take the perspective that, parallel to the
quantificational cases which are our main focus here, (a) modal operators are not anaphora barriers,
and (b) anaphoric accessibility is also controlled by whether a discourse is coherent.
5