Computational stylometry: predicting the authorship of investment arbitration awards

Abstract

The authorship of judicial opinions was an early target of computational legal studies (Oldfather et al., 2012). Principally focused on the US Supreme Court, different methods were deployed to identify the role of "unseen actors" in writing opinions. Motivated normatively by concerns about the disproportionate influence of clerks in "judicial ghostwriting" (Rosenthal and Yoon, 2011) and theoretically by an interest in the allocation and optimization of time in judicial labour (Choi and Gulati, 2005), research has sought to detect who are the authors of judgments, examine how they are influenced by unseen actors, and chart what variation exists across judges, actors and time.The methodology is grounded in stylometry, a linguistic theory premised on human authors leaving fingerprints in their writing and a method based on the statistical analysis of measurable patterns particular to literary style. Early stylometry focused on the prominence of "function words" (such as "a", "by", "if", "of" "their" and "who"), which were viewed as the most discernible textual fingerprint. The method permitted ground-breaking qualitative research in identifying authorship, among other things, of the Federalist Papers (Mosteller and Wallace, 1964), by focusing on the use of and placing of 63 function words. Others have sought to identify the authors of books of the Christian Bible (Morton and McLeman, 1966;Burns, 2006) and Shakespearean plays (Seletsky et al., 2007).This "supervised" approach to stylometry is backed by neurological studies, which show that preferences for function words develop late in children and that function and content words are stored and processed in different brain regions (Menn and Obler, 1990;King and Kutas, 1995). However, in stylometry theory, emphasis is also placed on features like word and sentence length, lexical frequencies, punctuation patterns, rare

Computational Methods in Legal Analysis

“…For example, Carlson et al (2016) conduct a stylometric analysis of US Supreme Court decisions based on term-frequency vectors of only function words, which are nonsemantic words (such as if and or). Frankenreiter (2019) and Langford et al (2020) extend this approach to the European Court of Justice and investment arbitration awards, respectively. The Linguistic Inquiry and Word Count tool offers various ways to summarize documents based on word frequency, for instance, whether there is a high occurrence of intensifiers such as the word clearly, and has been used to study language in court documents ).…”