Mixed-depth representations for natural language text
Graeme Hirst, Mark Dermot Ryan · 1992
Text understanding is usually hard for computers and easy for people, so we tend to forget about the times when it's hard for people, too. But even smart, knowledge-based people, and not just dumb computers, can nd text understanding extremely di cult. We all know this from our experiences with writing that presents complex ideas|advanced technical papers, for example| and writing that's just plain bad|incomprehensible instructions for assembling a Christmas toy, textbooks that present ideas sloppily, government tax-return guides that try hard to be clear but never quite succeed. 1 So it's no shame if a natural language understanding program, like ahuman, has to occasionally capitulate and say, in e ect, that it cannot fully understand some di cult piece of text. Now, intelligent text-based systems will vary as to the degree of di culty of the texts they deal with. Some may have a relatively easy time with texts for which fairly super cial processes will get useful results, such as, say, The New York Times or Julia Child's Favorite Recipes. But many systems will have towork on more di cult texts. Often, it is the complexity of the text that makes the system desirable in the rst place. It is for such systems that we need to think about making the deeper methods that are already studied in AI and computational linguistics