Note that term w3 has a non-zero probability estimate, even though it did not occur in the document text. If we add these three probabilities, we get which confirms that the probabilities are consistent. Different forms of estimation result from specifying the value of αD. The simplest choice would be to set it to a constant, i.e., αD = λ. The collection language model probability estimate we use for word qi is cqi/|C|, where cqi is the number of times a query word occurs in the collection of documents, and |C| is the total number of word occurrences in the collection. This gives us an estimate for P(qi|D) of: