Integrating Discourse Markers into a Pipelined Natural Language Generation Architecture

Integrating Discourse Markers into a Pipelined Natural Language Generation Architecture Charles B. Callaway ITC-irst, Trento, Italy via Sommarive, 18 Povo (Trento), Italy, I-38050 callaway@itc.it Abstract Pipelined Natural Language Generation (NLG) systems have grown increasingly complex as architectural modules were added to support language functionali- ties such as referring expressions, lexical choice, and revision. This has given rise to discussions about the relative placement of these new modules in the overall archi- tecture. Recent work on another aspect of multi-paragraph text, discourse mark- ers, indicates it is time to consider where a discourse marker insertion algorithm fits in. We present examples which suggest that in a pipelined NLG architecture, the best approach is to strongly tie it to a revi- sion component. Finally, we evaluate the approach in a working multi-page system. 1 Introduction Historically, work on NLG architecture has focused on integrating major disparate architectural modules such as discourse and sentence planners and sur- face realizers. More recently, as it was discovered that these components by themselves did not cre- ate highly readable prose, new types of architectural modules were introduced to deal with newly desired linguistic phenomena such as referring expressions, lexical choice, revision, and pronominalization. Adding each new module typically entailed that an NLG system designer would justify not only the reason for including the new module (i.e., what lin- guistic phenomena it produced that had been pre- viously unattainable) but how it was integrated into their architecture and why its placement was reason- ably optimal (cf., (Elhadad et al., 1997), pp. 4–7). At the same time, (Reiter, 1994) argued that im- plemented NLG systems were converging toward a de facto pipelined architecture (Figure 1) with minimal-to-nonexistent feedback between modules. Although several NLG architectures were pro- posed in opposition to such a linear arrangement (Kantrowitz and Bates, 1992; Cline, 1994), these re- search projects have not continued while pipelined architectures are still actively being pursued. In addition, Reiter concludes that although com- plete integration of architectural components is the- oretically a good idea, in practical engineering terms such a system would be too inefficient to operate and too complex to actually implement. Significantly, Reiter states that fully interconnecting every module would entail constructing N (N � 1) interfaces be- tween them. As the number of modules rises (i.e., as the number of large-scale features an NLG engineer wants to implement rises) the implementation cost rises exponentially. Moreover, this cost does not in- clude modifications that are not component specific, such as multilingualism. As text planners scale up to produce ever larger texts, the switch to multi-page prose will introduce new features, and consequentially the number of architectural modules will increase. For example, Mooney’s EEG system (Mooney, 1994), which cre- ated a full-page description of the Three-Mile Island nuclear plant disaster, contains components for dis- course knowledge, discourse organization, rhetori- Figure 1: A Typical Pipelined NLG Architecture cal relation structuring, sentence planning, and sur- face realization. Similarly, the STORYBOOK system (Callaway and Lester, 2002), which generated 2 to 3 …

Content truncated to 500 words