Abstract

Text written to inform often contains deixis to communicative artifacts such as textual structures (e.g., sections and lists), discourse entities, and illustrations. By relating such artifacts to the prose, “artifact deixis” plays an essential role in structuring the flow of information. In this talk I will describe ongoing work toward the goal of automatically detecting artifact deixis and its referents. First I will situate the phenomenon with related metalinguistic phenomena, applying prior work to its characterization. Next, I will describe recent results of a study of artifact deixis using a corpus rich in determiner-established instances of the phenomenon (e.g., “this section”, “these equations”, “those reasons”) sourced from Wikibooks, a collection of learning texts. This corpus is used in combination with WordNet to determine a set of word senses that are characteristic of the phenomenon, showing its diversity and validating intuitions about its qualities. The results motivate further research to extract the connections encoded by such deixis, with the goals of enhancing tools to present pedagogical e-texts to readers and, more broadly, improving language technologies that rely on deictic phenomena.