1. The document describes a patent application for phrase-based indexing in information retrieval systems. It involves identifying phrases in documents, indexing documents based on these phrases, ranking documents based on phrase matching, and using phrases to generate document descriptions. 2. Phrases are identified based on their ability to predict other related phrases. Documents are indexed with lists of the phrases they contain. Ranking considers how well document phrases match query phrases. 3. The system can identify related phrases and extensions when searching, detect duplicate and spam documents, and generate snippets for search results using highly ranked sentences.