The statistical approach is most helpful when we have large masses of data to analyze. The quantity of verbiage spoken daily by the average citizen is so great that detailed analysis of his every utterance is almost unthinkable. Nothing less powerful than the simplifying abstractions of statistics can deal with the data. What do people say? Even after we specify which people and under what conditions, the question is a hard one. It is not that we cannot record what they say, or that we cannot understand what they say, but that they say so much. Statistical simplification is imperative. We might make up a list that contained all the sentences we heard from the particular people under the given circumstances and check the appropriate sentence each time it occurred. If we did this for long, we should discover that some sentences occur relatively often, while others are rare, and some combinations of words do not occur at all. Before we begin to consider some of the particular restraints imposed upon what people say, however, we need a general statistical orientation to tell us what kinds of utterances are important to consider. One kind of constraint on the talker is the structure of the language he speaks. The successive words in an English sentence are not unrelated. Once a speaker has begun, 'She looked around the...,' he is not free to open the dictionary at random to select the next word. The next word is determined, at least in part, by the context of the words preceding it. It is this notion of context, of interdependencies among successive items in a message, that we must examine. Obviously we cannot study every verbal context that could conceivably occur. What we can do is draw a statistical map of our communicative behavior and ask why we permit ourselves so little freedom in the choice of what we say. (PsycINFO Database Record (c) 2012 APA, all rights reserved)