Higher-orderness and the richer control facilities are especially important for practical NLP systems. It is particularly desirable to be able to control search without sacrificing high-level and the essentially declarative description of data. Higher-order constructs are also used directly in current linguistic semantic formalisms.
If we distinguish between the inference engine and the linguistic workbench (i.e the grammar development environment) then richer control and compilation techniques are of the utmost importance for an efficient implementation of a particular grammar formalism. The object-oriented features might be useful for the implementation of the linguistic workbench itself.
Better software engineering facilities are essential to the NLP undertaking; even if most linguists do not consider themselves to be "programmers", they are still subject to the same complexity-control problems faced by the rest of the software industry. Indeed, the need for this kind of separation of concerns places even higher demands on modularity and abstraction facilities.
In sum: these are the days of real engineering, and if the open problems in NLP are construed to include what is necessary to its practical application, then everything becomes relevant. Indeed, such "peripheral" issues as user interface facilities can make or break a system - though this does not necessarily imply that they must be provided as primitives by the system.