Corpus linguistics for safeguarding children online

Safeguarding children in schools broadly refers to the actions taken to protect children from abuse, prevent damage to health and development, and promote conditions that would improve the life chances of children. To safeguard children, UK schools must implement filtering and monitoring software to “block harmful and inappropriate content without unreasonably impacting teaching and learning” (Department for Education, 2024: 40). The industry standard method for monitoring online language use in schools is ‘keyword monitoring’, which identifies the use or presence of specific words or phrases (e.g. ‘bomb’) that correlate with a specific form of risk (e.g. violence). However, this approach typically depends on lists of words isolated from their context(s) of use and tends only to raise concerns if there is a direct match to a ‘keyword’. This can lead to ‘false positives’ whereby a 'keyword' match raises an automatic safeguarding concern (e.g. ‘bomb’) even if the use of the keyword was innocuous (e.g. ‘bath bomb’). This paper introduces corpus linguistics as a set of methods and approaches to enhance the effectiveness of filtering and monitoring through a case study based on a 1094,914-word corpus of online testimonies relating to suicide. In doing so, we demonstrate how corpus methods and analysis of authentic language data can be used to identify and contextualise safeguarding concerns. The practical applications of this research are intended to help schools to better protect children from the illegal and legal (but harmful) online materials that currently pose a threat to their safety and wellbeing.