Handling Dependencies in Reorganizing Content Specifications – A Case Study of Case Analysis

Abstract

Handling the dependencies among alternatives in composing expressions in an efficient and qualitatively accurate manner is a fundamental problem of NLG. To pursue this goal effectively, simplifications are put forward in practical approaches, but also ambitious control regimes are tried out occasionally. However, neither of these is able to operate adequately on larger and involved structures. Approaching this issue in a methodological way, we present a case study from the area of mathematical proofs that illustrates the rhetorically motivated reorganization of machine-generated case analyses. Ingredients of this investigation are the design of optimization operations, orderings on groups of operations that take their dependencies into account, and tentative applications of local operations to test the effects of crucial dependencies. Our approach conceives NLG as a standard pipe-line architecture putting emphasis on orderings, with local revisions as a minor extension. This is particularly effective when text planning is organized as an optimization rather than as a construction process, such as for the presentation of mathematical proofs.

Formal Issues in Natural Language Generation

“…Helmut Horacek's paper illustrates the second idea, of integrating an NLG system with a formal reasoning engine (Horacek, 2004). In Horacek's case the engine is a theorem prover which provides the input to Horacek's system, which generates an English presentation of a proof created by the theorem prover.…”