This paper discusses some features of statistical approximations to English. It argues that the notion 'order of approximation' is unsatisfactory since it is not related to any objective characteristics of a string of words, and defines only a lower limit of semantic and syntactic cohesion. 'Contextual Constraint' is also discussed, and it is argued that this is not a single factor, but a complex series of interrelated factors with no obvious connection with length of context. The final section discusses statistical approximations to French, German and Welsh, and a number of types of constraint that are found in these languages but not in English. It is concluded that Miller and Selfridge's (1950) standard findings may not be generalizable to statistical approximations in languages other than English.