This corpus-based study deals with the process whereby new compounds become established as conventionalized lexical items in the Swedish language. Special attention is given to how their meanings develop during the early phases of usage. Strictly compositional accounts of their semantics are rejected. Nine recently conventionalized compounds have been investigated in detail. The main corpus consists of all citations of these compounds from two full-text databases consisting of newspapers and magazines. Together they cover almost all of the major newspapers and several other ones as well. In addition to these corpus data, a questionnaire was used where informants were asked to describe (what they thought was) the meaning of these compounds and also to indicate how sure they were about the conventional usage of these words. Furthermore, the Swedish Language Bank has been used, mainly to get frequency data which show what kinds of word formation patterns the component parts of the compounds usually take part in, and how strong these patterns are. These data are used as reference points when analyzing the actual uses of the compounds and the semantic associations that the informants had.

The compounds do not show a great deal of semantic variation. All but one has one clearly dominating sense, but most of the compounds are occasionally also used with strikingly deviant meanings. Metacomments and metasignalling seem to have an impact on conventionalization, at least they affect the familiarity of these items. While some of the compounds are extensively metacommented, explained and problematized, others are treated as natural and self-explanatory. The latter are generally found to instantiate strong word formation preferences, which make them seem transparent.