Authorship Identification: from Fundamental Limits to Practice

Authorship identification is the problem of estimating the author of a written text from a set of possible authors using clues from the written text itself. While this problem has been around for over 130 years, there yet remain several important questions regarding the fundamental limits of existing techniques. This paper is the first of our knowledge to build an information-theoretic framework around the problem of authorship identification. We present a model by which the limits of the problem may be studied as well as key definitions and metrics rooted in information theory. We also provide a test case from the historical archives of the authorship identification research, namely the Federalist Papers problem, and apply our methods to this well-studied set of texts. Given the adhoc nature of much of the research surrounding authorship identification (and related) problems, and the wealth of open fundamental questions regarding even modern methods based in machine learning and large language models, we anticipate that this work will lead to a number of future contributions in both authorship attribution and basic machine learning.