This paper presents a status report on Dragon Systems' large vocabulary, speaker-adaptive, continuous speech recognition research. Major system components and their algorithmic approaches will be described, as well as application tasks and performance on these. At present, vocabularies of greater than 1000 words up to 5000 words simultaneously active are running non-realtime on a 33 MHz 486 personal computer. Natural language tasks used range from highly specialized domains, such as mammography transcription (perplexity about 60), and general radiology (perplexity about 140), to broad encyclopedia dictation (perplexity about 430). Specialized tools created for development and graphical error analysis will also be described.