Developing corpus approaches to safeguarding and family justice system research.

Lead Research Organisation: Lancaster University
Department Name: Linguistics and English Language

Abstract

The research will create resources and tools to address contemporary safeguarding issues. Supportive networks around children help to ensure their wellbeing and mitigate threats to their safety from within and outside the family. Threats within families are the responsibility of Local Authorities, supported by professionals monitoring children for signs of abuse using risk prediction tools. Outside families, online threats to children are increasing and difficult to mitigate, including online bullying, grooming, pressure to send indecent photos, pornography, radicalisation, hate-speech, and fake news. Adults are increasingly expected to keep children safe online, detecting and heading off problems before they escalate. For disadvantaged families or those not digitally literate, this task can be daunting. Schools are similarly challenged in how they tackle children's online safety. Once in formal safeguarding processes, vulnerable families' information, advocacy and resources of a consistent standard, often from unregulated online sources. This project will create new linguistic resources and tools to support this.

Our project includes 'grey' safeguarding (online safety and risk prediction), as well as stages involving formal safeguarding interventions. Safeguarding applies to all children and considers children's safety within families, as well as safety in school, and in social contexts. Once children are identified as unable to be safeguarded without intervention, the family justice system (FJS) becomes involved. Safeguarding issues escalate into child protection processes when children are considered at risk of significant harm, and these processes can travel through the complex stages of the FJS. Cases that cannot be safely resolved for children progress to court, where children can be removed from their families, sometimes permanently. Increasing numbers of children in England are escalating through this system, causing strain on the courts. Contributing to this are the many families who cannot afford legal advice and who represent themselves in court.

Four significant areas of concern have been identified in previous FJS research. First, increasing numbers of people cannot afford to pay for the advice and representation they need. This is harming families and causing problems for systems managing increasing numbers of vulnerable, unsupported litigants in person appearing before judges without an advocate. Second, the increasing use of unregulated and often unreliable predictive technologies, generating risk scores which have proved to be opaque and can be unreliable. Third, schools and parents are tasked with keeping children safe online, using technologies in their infancy to monitor children's internet activity and identify issues. Fourth, in cases that progress to court, predictive expert evidence is important, but such evidence can be unreliable and no large-scale study of its contents has taken place prior to this project.

Our project provides research and new corpora to help resolve these issues, contextualised by an analysis of comparable global systems to inform suggested solutions. We combine two established fields of research for the first time, corpus linguistics/corpus-assisted analysis, and safeguarding/family justice system research (which has not significantly drawn on corpus methods before). Our project will carefully create new corpora (searchable, text datasets), and new tools from our large volumes of combined data. This data will be used for our analyses before being made available for future projects. This is the first time a volume of detailed case level datasets extending to millions of words will be able to be used for both quantitative and qualitative analysis. This is made possible by our development of a unique, semi-automated, anonymisation protocol which we will apply to the data. Our approach will enable statistically reliable findings, extending beyond the high-level statistics.

Publications

10 25 50