Well-known examples such as the following show that a single phrase can appear to simultaneously fulfill conflicting syntactic requirements:

   Kogo Janek lubi    a   Jerzy nienawidzi (Polish)
   who  Janek likes(OBJ=ACC)   and Jerzy hates (OBJ=GEN)
   `Who does Janek like and Jerzy hate?'

The noun phrase `kogo' is the object of two coordinated verbs; the first verb requires that its object bears accusative case, while the second requires genitive case. `Kogo' can fill both of these requirements, as the accusative and genitive forms for this particular noun are the same. It is not possible to characterize `kogo' as bearing either accusative or genitive case, since it must satisfy both requirements simultaneously.

We analyze such examples by making use of the fact that LFG allows feature values to be sets of items as well as individual symbols. The CASE value for an ordinary genitive noun is represented as the singleton set {GEN} instead of the symbol GEN, and an accusative noun has the value {ACC}. A word that can satisfy both genitive and accusative requirements has {ACC,GEN} as its CASE value, where the different elements indicate the different possibilities of realization. Finally, the contextual case-agreement requirements are imposed by set-membership assertions instead of equality conditions that would lead to a contradiction.

Set representations can also provide an analysis of feature resolution in coordination: how certain features of conjoined elements are resolved to determine the agreement features of the coordination as a whole. For example, a coordinate noun phrase with a first person conjunct behaves as a first person plural form; analogously, a Hindi or French coordinate noun phrase with a masculine conjunct behaves as a masculine form. We propose that the PERSON and GENDER features have complex values which we will represent as sets of more primitive markers; we assign set values to features of the individual conjuncts, and take the union of the conjunct sets to be the feature value for the coordinate structure.

We present our solution within the formalism of Lexical Functional Grammar, but we presume that similar techniques could be applied in other description-based frameworks.