A comment on Alison Finch's article - comment on article by Alison Finch in this issue, p. 511

Style, Winter, 1995 by Louis T. Milic

Alison Finch's article ("Computer-Assisted Research on Literature: The Imagery of a Myth") is touching in its reliance on old-fashioned scrutiny of metaphors and similes and her attempt to build out of these a structure of what she calls a myth. Unfortunately, the points she makes are in large part either irrelevant or wrong. She cites the claim that "almost all literary critics of repute" ignore the results of computer-assisted research of literature, that this research is presented in a "rebarbative" way, and - this is the main point of her article - something that she claims has not been noticed by any critics but herself, that figures of speech used by practitioners in presenting their research "tend to mythologize their own enterprises." In order to support this latter claim she has put together a farrago of citations from many different sources, without concern for context, meaning, or anything else.

There is no doubt that some (especially Francophone) researchers have gone far afield in presenting their data and their conclusions, but Finch's claim is not only foolish, but also misguided. She does not seem to realize what computer assisted research is and what it does. To begin with, research in literature (computer-assisted or otherwise) is essentially about language. Literature is a sub-field of the mass of language of all kinds that is constantly being produced by its speakers. Hence the study of literature, as well as the study of newspapers, law books, and children's books, is the province of linguistics, specifically corpus linguistics, which has a broad constituency, of which Finch has apparently never heard. This is what computer-assisted research really is. Computer study of language is essentially a quantitative discipline and as such is subject to the methods and principles of statistics.

Computer corpora originated in this country, with the publication in 1961 of the so-called Brown Corpus, which has been extensively analyzed by researchers. British and Scandinavian scholars have been developing their own corpora (The Lancaster-Oslo-Bergen, London-Lund, Helsinki Corpus, British National Corpus [in progress]), analyzing them, and publishing their results in, first, the I.C.A.M.E. Journal and, later, in other outlets (see volumes by Stig Johansson and Anna-Brita Stenstrom, Jan Svartvik, Gerhard Leitner, and Merja Kyto, Matti Rissanen, and Susan Wright, inter alia), and have established computer-linguistic analysis on a solid basis, constantly exfoliating (to use an arboreal metaphor).

The major difference between these activities and those of critics applying a package of software to a poem, some short stories, or the like, is that proper computer study requires a substantial mass of text and suitable comparative norms. In an earlier age (the late nineteenth century and the beginning decades of the present one) scholars were doing manual counts and producing interesting results. The case of Lucius Sherman of the University of Nebraska is instructive: This scholar found his classes in literature barren compared to those of his scientific colleagues and proposed to put the study of language - he said literature, but he meant language - on a scientific basis. In his book Analytics of Literature, he surveys sentence-length over a period of five centuries, from Fabyan the historian to Macaulay. Counting sentence-length by hand, he found a regular average decline over the period, when he counted at least 500 sentences. He went as far as 2255 sentences of De Quincey and found that this imaginative writer consistently averaged 33 words per sentence. He went even further with Macaulay, all of whose Essays and his History of England he counted (presumably with the assistance of his students) and discovered that the average sentence length of 23.50 words is absolutely constant, however many sentences were in his sample.

Although Sherman was a beginner in statistics and an amateur in science, he stumbled on some interesting facts in his experiments, facts which are still useful today. Similar results can be found in the work of the researchers in corpora and some few of those who have dipped into literature, though the latter have often been hampered by a lack of background in statistics, programming, and scientific method. Computer-assisted research in language is not defective because of its figures of speech, but only as it does not follow suitable principles of quantification. Alison Finch's article, while amusing in its presentation of citations, is naive in its lack of understanding of the sort of research she is attacking. Despite this judgment, I concede that Finch's essay should be published. It represents a view of computer-assisted analysis of style that is far too prominent, but publishing it with my reply allows readers of the journal to see the two views in stark contrast to one another and to evaluate the claims of each.

Works Cited

Johansson, Stig, and Anna-Brita Stenstrom. English Computer Corpora. Berlin: Mouton de Gruyter, 1991.