From negative statements to positive safety

The GB railway industry maintains a database of close calls which are “hazardous situations where the event sequence could lead to an accident if it had not been interrupted by a planned intervention or by a random event” (Gnoni et al. 2013). The data is entered as free-text by railway workers, however the very large number of events reported each year makes it infeasible for every record to be manually read to obtain safety information. Computer-assisted natural language processing (NLP) techniques can be employed to assist in reading the records. Whilst the current NLP literature describe a range of techniques to extract information, no overall approach has been identified that is effective in every situation: Grimmer & Stewart (2013) state that specific processes need to applied to solve specific problems.