Aggregation in Natural Language Generation: Another Look

Aggregation is gaining attention as an important phenomenon in automated natural language generation. The small quantity of previous work which deals with this subject suuers from some conceptual diiculties, including inadequate deenition (i.e., as a process which eliminates redundancy), vagueness as to its locus in the generation process, and unclear relation to other phenomena. It is concluded here that aggregation is best seen as an emergent phenomenon which results from careful sentence planning. This view draws attention to the processes of semantic grouping, sentence structuring, and sentence delim-itation which together underlie aggregation eeects. Within the framework of the HealthDoc project, an attempt to develop systems for automatic patient text customization, the application of the above perspective is considered.