Page 394 - Encyclopedia of Philosophy of Language

Formal Semantics
ral-language grammar, or are they rather part of some general model structure onto which natural-language strings are mapped. A formal algorithm is denned from natural language expressions onto DRstructures, suggesting that they are envisaged as being pan of natural-language grammars. Yet they are defined for entities for which those grammars do not provide input—viz., sequences of sentences. This is the prob- lem which ties at the heart of the context-dependency phenomenon: the articulation of truth-evaluable vehicles is not defined over sentence-strings as deter- mined by principles internal to the grammar alone, but by such principles in combination with something else. If we are to isolate a concept of natural language content attributable to natural language expressions independent of context, then we have to model two related phenomena: (a) the underdeterminacy dis- played by many natural language expressions, simple and complex, vis-a-vis the truth-theoretic content attributable to them; and (b) the process whereby the lexically-assigned natural-language content is enriched to yield some such complete specification. DRT was one of the first theories of semantic modeling to seriously grapple with this problem (cf. also Heim 1982; Barwise and Perry 1984).
The context-dependency of truth-theoretic content is no less problematic for standard syntactic assump- tions, for the recognition that grammar-internalspeci- fication of anaphoric expressions very considerably underdetermines interpretation conflicts with familiar syntactic distinctions. There is claimed to be a dis- tinction between discourse coreference and bound variable anaphora, the latter (but not the former) being subject to a c-command restriction between it and the operator on which it is dependent (Reinhart 1983, 1986). Hence the widely adopted grammar- internalcharacterization ofbound-variable anaphora. This assumption that some anaphoric dependencies are determined exclusively by grammar-internal prin- ciples cannot be sustained. The reason is this: for every anaphoric linkage, howsoever established, there is a corresponding bound-variable analogue. In particular there are anaphoric linkages demonstrating two major central cognitive activities:
(a) logical deduction
Joan isn't so anti-private practice as not to have (19) any private patients, but she's always
complaining that they treat her as a servant.
(b) retrieval from memory of contingently known information associated with specified objects
The fridge is broken. The door needs mending. (20)
Establishing the anaphoric linkage in (19) involves a step of'double negation elimination': establishing the anaphoric linkage in (20) involves making a link in virtue of the knowledge that fridges have doors. Both processes are central to any account of human reason-
ing of the most general sort, and are not properties of the language faculty itself. But these examples have straightforward analogues in which the pronominal linkage involves central cognitive processes while yet licensing a bound-variable interpretation:
Every one of my friends who isn't so ami- (21) private practice as not to have any private
patients is complaining that they treat her as a
servant.
Every fridge needs the door mending. (22)
These data pose us with a number of alternatives, only one of which is free of inconsistency. They display an interaction precluded by all standard theories of syntax, that between general cognitive processes and constraints said to be subject to grammar-internal explication. It appears that the output of such general cognitive processes has to be checked against a syn- tactic restriction on interpretation. We could take these data as evidence that the encapsulation of the language faculty should simply be abandoned and free interaction of processes internal to the language faculty and central cognitive processes should be allowed. Since this would involve jettisoning all possi- bility of characterizing properties specific to language, this alternative is not acceptable. Notice however that the alternative of invoking the ambiguity of bound- variable anaphora and discourse coreference as a means of dividing off grammar-internal processes from general cognitive processes is not a viable option. To postulate ambiguity here is no help—the grammar- internal phenomenon still involves interaction with the relevant central cognitive processes. And to stipu- late double negation elimination or bridging cross- reference as a grammar-internal phenomenon is to incorporate central cognitive processes into the gram- mar, and this too involves reneging on the language- encapsulation view. Our only recourse is to grant the underdeterminacy of all anaphoric expressions, allow the phenomenon of anaphoric dependence to be char- acterized as part of the pragmatic process of assigning interpretation to utterances, and characterize the con- straint on bound-variable interpretations in like man- ner to the disjointness requirement on pronominal
anaphora (principle B) as a filter on licit choices of anaphoric dependence made as part of this pragmatic process.
Exploring this last route gives us a much more syn- tactic view of content. We need to define concepts of locality, c-command, etc., over configurations licensed both by grammar-internal processes and by general cognitive processes. There is evidence that this is the right direction in which to look for a solution. Ellip- tical processes display the underdeterminacy of natu- ral-language expressions vis-a-vis the interpretation assigned to them even more dramatically than anaphora, but are yet subject to familiar grammar- internal constraints such as the so-called 'island' con-
372