On the role of parole candidates' language in parole board hearings

Kurzfassung/Abstract

Granting parole is viewed as a critical element of criminal justice for parole candidates. An emerging stream of research investigates the drivers of parole board decisions with respect to granting versus denying parole, the timing of parole and also recidivism of parole candidates released on parole. Procedurally, a parole suitability hearing is a verbal exchange between the parole board and the parole candidate. In a sense, this hearing provides candidates the opportunity to “present their case” for parole and for the board to obtain information about the candidate that is not available from their case.
Given that the exchange in hearings is verbal, we investigate the influence of the language used by candidates in parole suitability hearings on parole board decisions. We harness the power of large language models such as OpenAI's GPT series of models to augment a variety of characteristics from their speech. We use these characteristics in a model to predict parole board decisions and find that their influence on decisions is significant. We also find that accounting for parole candidate's speech changes the role of other variables (crime, race) which suggests that standard parole prediction models miss a fundamental element of the parole decision making process. Implications and directions for future research and practice are discussed.