Constraints on Lexical NP Selection

Constraints on Lexical NP Selection
Referential NP generation is the task of building a description for an object that is known
to the generator, under the assumption that hte hearer wants to know who the referent
is. The received wisdom is that there is a hierarchy of NPs that one could use for this
task starting with the local pronouns and the reflexive and reciprocal pronouns, going down
through third person pronouns, demonstrative pronouns, names, demonstrative lexical NPs,
definite descriptions down to indefinite descrptions. The task is a question of walking down
the hierarchy and to see whether the NP type in question is appropriate: if it works, the
choice is right. If it does not work, one takes a step down in the hierarchy.
The hierarchy is related to the quality of the identification, the ease of identification and
to the power supplied to the hearer by the identification. The quality can be described as
the reliability of the identification, ease as the amount of effort that the hearer needs to
put into the identification and power as the handle that the hearer will get on the referent.
Local pronouns give a very high degree of all three properties: for singular local pronouns,
identification is almost guaranteed, the referents are higly activated, and the referents are
given in the speech situation where the hearer is observing them, address them or carry out
other action. For indefinites the three dimensions are low: effort is high since the referent
is new, it is only the full clause that gives a unique description, and that description is not
guaranteed to enable any action on the referent.
The whole hierarchy guarantees identification. Certain uses of definite descriptions do not
have the uniqueness property but they share with referential indefinite descriptions the property that a unique description is provided by the nominal material together with the rest of
the clause (the Evans property). The talk is about the lexical NPs and explores the hypothesis that the same three concerns give the preferences in the choice of lexical material and the
choice for the article. The speaker chooses the material and the article, guided by his take on
the hearer strategy and resources.
Non-unique descriptions with the definite article would be prefered over other descriptions
because of ease and power. They are evaluated by direct association to elements in the
common ground which serve as antecedent, as origin of a bridge or as indirect indicator of the
referent. This requires little effort and the antecedent offers the best possible handle on the
referent. Bridging does not guarantee uniqueness. Non-unique definite descriptions are not
able to accommodate in the strict sense. Unique descriptions require the definite article, and
freely allow accommodation. Accommodating descriptions are distinguished from indefinite
descriptions in giving more power by allowing at least two full strategies of identification: from
the main eventuality in the clause and from the description itself. The hypothesis predicts a
preference for nonunique descriptions that identify the referent directly over bridging uses by
ease. And of bridging uses over unique descriptions.
The Evans property entails that in choosing lexical material for an indefinite reference, identity
is still a consideration. The full clause must yield a unique description of the referent. But
bridging is also central for the choice of the description in indefinites. A bridge or a subsection
increases the power of the hearer on the referent and if possible should have preference over
descriptions that do not have one. Both givenness and uniqueness are obligatory marked by
a definte article and while they often coincide, they need not.