Mental profile mapping: A psychological single-candidate authorship attribution method.

Ryan L Boyd,Zhiqiang Cai

doi:10.1371/journal.pone.0200588

Abstract

Modern authorship attribution methods are often comprised of powerful yet opaque machine learning algorithms. While much of this work lends itself to concrete outcomes in the form of probability scores, advanced approaches typically preclude deeper insights in the form of psychological interpretation. Additionally, few attribution methods exist for single-candidate authorship problems, most of which require large amounts of supplemental data to perform and none of which rely upon explicitly psychological measures. The current study introduces Mental Profile Mapping, a new authorship attribution technique for single-candidate authorship questions that is founded on previous scientific research pertaining to the nature of language and psychology. In the current study, baseline expectations for results and performance are set using an advanced technique known as “unmasking” on the test case of Aphra Behn, a 17th century English playwright. Following this, Mental Profile Mapping is introduced and tested for its psychometric properties, tested using a “bogus insertion” method, and then applied to canonical Aphra Behn plays. Results from both attribution methods suggest that 2 of 5 questioned plays are likely to have been authored by Behn, with the remaining 3 plays exhibiting a poor fit for Behn’s psychological fingerprint. Mental Profile Mapping results are then decomposed into deeper psychological interpretation, a quality unique to this new method.

Highlights

Authorship attribution is, broadly speaking, the process by which works of unknown or disputed origins are investigated to determine their history
The past 2 decades have seen an explosion of new methods in the world of authorship attribution, those that employ the statistical modeling of language to determine authorship likelihoods [1]
Despite the recent boom in sophisticated text analytic authorship attribution methods, tensions often exist in forensic settings where impenetrable algorithms are given free reign over authorship questions to the exclusion of intuitive, digestible, and human insights [2]

Summary

Introduction

Authorship attribution is, broadly speaking, the process by which works of unknown or disputed origins are investigated to determine their history. The past 2 decades have seen an explosion of new methods in the world of authorship attribution, those that employ the statistical modeling of language to determine authorship likelihoods [1]. Despite the recent boom in sophisticated text analytic authorship attribution methods, tensions often exist in forensic settings where impenetrable algorithms are given free reign over authorship questions to the exclusion of intuitive, digestible, and human insights [2]. Complex machine-learning methods may jeopardize a layperson’s ability to interpret the results of forensic text analyses, and the ability of expert researchers themselves to adequately understand the process by which results are attained (e.g., by interpreting an algorithm’s resultant model)

Methods

Results

Discussion

Conclusion