Words, frequencies, and texts (particularly Conrad): a stratified approach

This paper describes a corpus methodology developed in the course of investigations of three small subcorpora of texts: fiction by Conrad, nineteenth-century accounts of exploration, and adventure fiction. Stemming from observations that both high and lower frequency words in text are of interest in relation to meaning and structure, a 'stratified approach' integrates different levels of frequency by grouping words in a text into a series of frequency bands, identified from a reference corpus, then examining the most recurrent words within each band. This approach shows up, in particular, the ways in which topic and theme are lexicalized, and connotative lexis is associated with lower frequencies. In addition to exemplifying the approach and demonstrating its applications, the paper argues that any set of words, identified as significant through frequency, still involves selection, and that information which is derived quantitatively should not be interpreted as if wholly objective.