A system stores digital media records, and comprising a vocabulary file of words keyed to the digital media records. A search query is received by computer from a user, the search query including words. The search query is logged by computer, yielding a query log. The query log is processed by computer to identify words in the query log which are not words already included in the vocabulary file and which are not words which are variants of words already included in the vocabulary file. Such words are defined as "not-found words." The not-found words are added to the vocabulary file, and the not-found words are keyed to the digital media records.