1.4 Operations in the Data Model

The following table outlines some of the main operations that may be performed in an NLG system, together with their primary input and output types.

Operation Description Primary input Output

CON Content selection ConRep

RHET Rhetorical structuring ConRep AbsRhetRep

*AGG Aggregation * *

LEX First stage of lexicalisation ConRep SemRep

TLC True lexical choice SemRep AbsSynRep

REG Referring expression generation ConRep SemRep

REAL Surface realisation AbsSynRep SynRep

This table ignores extra ``contextual'' inputs that the operations may need. For instance, lexical choice may need access to a user model (see 1.5 below). In addition, a finer-grained notion of types is likely to be useful in several places in order to characterise the operations more precisely. For instance, it may need to be specified what aspects of abstract syntactic representation are determined by TLC. Some progress on this can probably now be made, given the detailed specification of some of the levels of representation later in this document.

Lexicalisation is assumed to take place in general in two stages, LEX and TLC [Cahill 1998]. Aggregation is considered to be a generic operation that is implemented by a family of operators, one for every level of representation [Reape 1999a]. This means that most levels of representation need to allow for the representation of ``aggregated structures''.