The best results reported here (e.g. row 8 of Table 4) rely on the use of BM11 with a query term frequency component from the ``long'' queries. Without a qtf component results from the long queries are not good (Table 1, bottom half). Even the ``short'' TREC-derived queries are very long compared with the searches entered by end-users of interactive retrieval systems. However, they are not long in comparison with user need statements elicited by human intermediaries. This suggests that it might be possible to make good use, even in an interactive system, of quite long and even rambling query statements, if users can be persuaded to enter them. We might even suggest a voice-recognition system!