Literary attribution and likelihood-ratio tests: The case of the middle EnglishPearl-poems

The kind of literary attribution problem best suited to statistical treatment is one in which two or more known individuals are the only candidates for the authorship of a disputed text. In such cases, analysis of valid stylistic variables in the known works of each author-candidate determines the statistics that can distinguish each one from the others. The measures of these variables in the disputed work provide a basis for predicting the known author to whose population of works the disputed text belongs. No doubt the classic approach to this kind of problem is that of Mosteller and Wallace, whose largely Bayesian analysis of the Federalist papers established a convincing case for attributing all of the thirteen disputed essays to Madison rather than to Hamilton.1 On the other hand, severe difficulties occur in an attribution problem in which only one candidate exists for an anonymous work. This is true, for example, of Milic's attempt to ascribe to Swift the disputed work A Letter of Advice to a Young Poet. z Since no convincing external evidence exists for including it in the Swift canon, this work has been attributed to Swift by some scholars and rejected by others. But no compelling rival to Swift exists, and hence as Matlack and Matlack point out in their review of Milic's study, unless " random samples of a possible non-Swiftian candidate can be specified," the ascribing of the work to Swift on the basis of a quantitative analysis of the style of the work "cannot be done. ''3 A different and unusual type of problem arises from the existence of two or more anonymous texts whose close association prompts us to ask whether they are the works of one or of many authors. A statistical solution to this kind of