This paper explores the effectiveness of surface text patterns in open-domain question-answering systems. The authors develop a method to automatically learn and standardize these patterns from the web, using a bootstrapping process with a few hand-crafted examples. The patterns are extracted from documents returned by AltaVista and their precision is calculated. The system then applies these patterns to answer new questions, with results tested on the TREC-10 corpus and web data. The study finds that the system performs better on web data due to the abundance of high-precision answers. However, the patterns have limitations, such as handling long-distance dependencies and multiple anchor points in questions. The paper concludes by suggesting the integration of web and TREC corpus outputs and the potential for multilingual QA using this method.This paper explores the effectiveness of surface text patterns in open-domain question-answering systems. The authors develop a method to automatically learn and standardize these patterns from the web, using a bootstrapping process with a few hand-crafted examples. The patterns are extracted from documents returned by AltaVista and their precision is calculated. The system then applies these patterns to answer new questions, with results tested on the TREC-10 corpus and web data. The study finds that the system performs better on web data due to the abundance of high-precision answers. However, the patterns have limitations, such as handling long-distance dependencies and multiple anchor points in questions. The paper concludes by suggesting the integration of web and TREC corpus outputs and the potential for multilingual QA using this method.