Abstract
In this paper we examine the "jumbled words" effect which denotes human ability to easily read words whose internal letters have been re-arranged as long as external letters remain in their positions. Hitherto, many explanations for this effect have focussed on the processes that operate "bottom-up". Here we suggest that "top-down" processes also play an important role and demonstrate this experimentally. First, we briefly describe the main types of word-recognition models and consider which model best explains the effect. Then, we present an experiment in which jumbled words of different frequency of occurrence were immersed in various types of contexts. Results indicate that both the frequency and semantic sentential context are involved in jumbled word recognition. The implications of these findings for word recognition models are discussed.