[hnbad]

Popular crime shows like CSI (and arguably detective novels even before that) have created a false impression that forensics are an exact science that can perfectly narrow down the list of suspects to the exact person who did it.

But of course in reality it is much more complex. All forms of forensic evidence are vulnerable to noise: sure, that one interesting artifact may be evidence but it also may not be, and the inverse may be true for that perfectly ordinary thing everyone missed. A linguistic quirk can be a piece of evidence but it can also be accidental (e.g. nowadays it might also result from bad predictive typing or autocomplete, or even more recently, as others have pointed out, LLMs).

So all of these are in effect just probabilistic filters. And filters are only useful when your sample size includes your target (i.e. if the actual perpetrator is a suspect and you have adequate data about them). And even then they may not only produce false positives but also false negatives and these may interact the more filters you attempt to combine.

Forensic linguistics can be useful when you have a small set of suspects that you absolutely know includes the actual perpetrator. But otherwise they can send you on a wild goose chase or hurt the innocent.

[psunavy03]

Trial lawyers have written about the "CSI Effect," where the existence of such shows produces jurors who now expect trials to contain the types of flashy scientific evidence they see on TV, and become less likely to convict even obviously-guilty people when this type of evidence can't be realistically produced.