The myth of bad passive voice and weak words an empirical investigation in the automotive industry

In requirements engineering literature dealing with natural language specifications, we usually find writing rules like `avoid passive voice' or `do not use weak words'. Adhering to such rules should result in understandable and unambiguous requirements. Passive voice, especially when used without an explicit actor, is considered to result in incomplete requirements. The usage of weak words is considered to result in imprecise requirements that are hardly testable. But is the inversion of the claim correct, i.e. does the violation of the writing rules result in problematic specifications? At least in our environment (the passenger car development of Mercedes-Benz) we observe that authors often use passive voice, and there are many requirements containing weak words. To answer this question, we conducted an empirical investigation whose results we report in this paper. The results of this investigation are quite surprising: The use of passive voice, even when the actor is missing, is almost never problematic, as the missing information (the actor) can in most cases easily derived from the context (i.e. surrounding requirements or the general project context). The usage of weak words may be considered problematic in approximately 12% of all occurrences. For an automatic analysis on weak words linguistic patterns can be defined to detect these problematic occurences.