To parse or not to parse

We have designed and implemented a text processing system that can extract important information from hundreds of paragraphs per hour and can be transported within weeks to a new domain. The system performs efficiently because it determines the level of processing required to understand a text. This "skimming" method identifies surface relations in the input text that are likely to contribute to its interpretation in a domain. This approach differs from previous skimming techniques in that it uses conceptual information as part of bottom-up linguistic processing, thus using linguistic knowledge more fully while limiting grammatical complexity.