Retrieving Customary Web Language to Assist Writers

This paper introduces NETSPEAK, a Web service which assists writers in finding adequate expressions. To provide statistically relevant suggestions, the service indexes more than 1.8 billion n-grams, n ≤ 5, along with their occurrence frequencies on the Web. If in doubt about a wording, a user can specify a query that has wildcards inserted at those positions where she feels uncertain. Queries define patterns for which a ranked list of matching n-grams along with usage examples are retrieved. The ranking reflects the occurrence frequencies of the n-grams and informs about both absolute and relative usage. Given this choice of customary wordings, one can easily select the most appropriate. Especially second-language speakers can learn about style conventions and language usage. To guarantee response times within milliseconds we have developed an index that considers occurrence probabilities, allowing for a biased sampling during retrieval. Our analysis shows that the extreme speedup obtained with this strategy (factor 68) comes without significant loss in retrieval quality. C. Gurrin et al (Eds.): Advances in Information Retrieval Proceedings of the 32nd European Conference on Information Retrieval, ECIR 2010 Milton Keynes, UK, pp. 631-635, ©Springer 2010.