Many discourse studies before the turn of the millennium were not corpus-based and did not use quantitative methods; consequently, their results were not generalizable. The reason for restricting analyses to smaller samples of texts was mainly the result of the lack of corpus analysis tools for automatic identification of various important textual features (such as lexical repetition, pronoun reference, and so on). The analysis of such features is still impossible to automate, as they require the reader’s background knowledge and comprehension of textual connections. Identifying the referent of a pronoun, for example, involves selecting the text segment (to any extent, from a word to a whole sentence) that represents the same entity in the universe of the particular discourse studied. This involves not only subject-specific background knowledge but also a complex decision-making process involving part-whole relationships between concepts. It is difficult to imagine any computer program that could establish such connections. The best that can be done is to automate some of the processes involved, such as coding or assessing results by interactive programs (Biber et al., 2005), which are becoming increasingly popular (e.g., WORDSTAT, LEXALYTICS, DICTION, WORDSMITH, or the Discourse Profiler, to mention a few). While they do have a few useful tools (such as type/token or other frequency counts), they are little help in the analysis of reference.