Since the seminal 1957 studies of word learning by Roger Brown, most experimental studies of lexical acquisition have concerned fast mapping: the process through which a new lexical entry is established, and through which representations of the linguistic context of a newly heard word interact with representations of its nonlinguistic context to fix an initial partial meaning. Here I focus on the subsequent extended process through which the adult meaning is approximated. Two factors lead to an extended learning process; the size of the hypothesis space and the need, sometimes, for the creation of new semantic primitives. Sometimes lexical learning requires conceptual change. I sketch a learning mechanism through which this can be achieved. A case study of learning the meanings of verbal numerals illustrates the argument.