Large Vocabulary Word Scoring (LVWS) integrates word spotting and Large Vocabulary Continuous Speech Recognition (LVCSR) to generate scores and times for a complete lexicon of several thousand words. We first discuss the main characteristic of LVWS and consider two implementations, one based on posterior probability scoring, the other on the n-best lists. Then we address the issue of how to map the scores to more meaningful confidence scores. We propose two solutions and present experimental results on the Switchboard corpus where we associate confidence scores to the words on the recognition output. Finally we discuss how to generate transcriptions directly from the LVWS output. We propose a simple approach and compare its performance to the performance of the LVCSR system.