Which vs. that: a corpus study

Based on corpora and earlier studies, this paper mainly attempts to answer the question how constructions of restrictive [i]which[/i] and [i]that[/i] have developed in comparison to one another in American and British English news until today. Corpus queries are designed to match patterns of particular object and subject gap constructions, such as [i]I like the ball that/which is green[/i] with subject gap and [i]It is the toy that/which I prefer[/i], which has object gap. Each query generates a [i]query set[/i], which includes all search hits for that query. Rather than checking all entries in every query set, entries of randomized samples are verified manually. From each such sample, the proportion of relevant entries, [i]relevance index[/i], is calculated. Relevance index helps us to estimate the relevant frequencies of the query sets. These estimations are essential for calculation of [i]frequency indexes[/i], which compare how frequencies of [i]that[/i] and [i]which[/i] clauses have progressed over time. In British English, the results are mixed with opposite tendencies for different time periods and news categories. As for American English, all data consistently support a significant frequency increase in [i]that[/i] with a corresponding decline in restrictive [i]which[/i].