Lightweight Structured Text Processing

Text is a popular storage and distribution format for information, partly due to generic text-processing tools like Unix grep and sort. Unfortunately, existing generic tools make assumptions about text format (e.g., each line is a record) that limit their applicability. Custom-built tools are one alternative, but they require substantial time investment and programming expertise. We describe a new approach, lightweight structured text processing, which overcomes these difficulties by enabling users to define text structure interactively and manipulate the structure with generic tools. Our prototype system, LAPIS, is a web browser that can highlight, filter, and sort text regions described by the user. LAPIS has several advantages over other systems: (1) the ability to define custom structure with a simple, intuitive pattern language; (2) interactive specification, showing pattern matches in context and letting users choose the most convenient combination of manual selection and pattern matching; and (3) external parsers for standard text formats. The pattern language in LAPIS, text constraints, describes text structure in high-level terms, with region relationships like before, after, in, and contains. We describe an implementation of text constraints using a novel, compact representation of region sets as collections of rectangles, or region intervals. We also illustrate some examples of applying LAPIS to web pages, text files, and source code. Appeared in Proceedings of USENIX 1999 Annual Technical Conference, Monterey, CA, June 1999, pp 131–144. Outstanding paper award.