A User- and Text-Oriented Approach to Annotation
Abstract
In the decades-long enterprise of information extraction (IE) system development, it has generally been presumed that systems for automatically identifying various categories of information in text are basically helpful to analysts as consumers of content leading to decision making (DM). Recent studies have shed light on the complexity of that account. Our research suggests that analysts probably benefit from different kinds of decision aids, depending on the genres of texts presented to them (e.g., tweets, narratives) and the types of questions they must, based on text content, answer (e.g., inferential, multiple choice). Here, we summarize education community research exploring this very space and present novel experiments involving different types of text (e.g., scientific expository texts, collections of military intelligence reports), asking different kinds of questions about the text (e.g., simple information queries, queries requiring multiple steps of deduction or inference), and providing different styles of annotation (e.g., user/computer generated, rich/sparse), as DM aids, to point up the relevance of these dimensions on the usefulness of annotation for improved text comprehension and problem solving by analysts and learners.
Document Details
- Document Type
- Technical Report
- Publication Date
- Sep 01, 2020
- Accession Number
- AD1109546
Entities
People
- Erin Zaroukian
- Michelle Vanni
- Sue E. Kase
Organizations
- United States Army Research Laboratory