Tim Grant on forensic linguistics

Tim Grant, Deputy Director at the Centre for Forensic Linguistics at Aston University has helped police on a number of cases, including the hate mail letters that been sent to addresses across England and Scotland. Here he talks about how his team's analysis of documents and writings can help police with their investigations.

What is academic linguistics?

Academic linguistics is the study and description of language patterns, both traditional grammatical descriptions and how we use language across a variety of contexts.

What is forensic linguistics?

Forensic linguistics is the description of language used in forensic contexts. Forensic situations include police interviews, courtrooms or writings that become part of an investigation. We don't only work on investigations; we're also interested in wills and contracts, the language police use asking questions in interviews and the language of lawyers and judges in courtrooms.

How is it helpful to the police?

It depends on the case. Sometimes we’re asked to look at writings in cases where police already have a suspect. In these cases we can compare the known writings of the suspect with the incriminating documents. The task then is to try to say whether the suspect wrote them.

Recent cases have included analyses of terrorist conspiracy documents, of a fraudulent suicide note and of a set of threatening emails. I've developed a particular interest in text messages and internet chat and have been involved in a number of cases in this area. Our evidence in these comparitative cases can be admitted to court.

In cases where police don't yet have a suspect, we can look at the language and try to work out the sort of person the writer might be. We may be able to say whether there’s more than one writer, whether they're a man or a woman and their likely background. This work is based more on likelihood than certainty but can be helpful in scaling down the search.

In other cases police know who wrote the text but might want to know what the person meant. For example, if I write in a text that I’m going to "juk" or "bore" you this can be street slang meaning to stab and could constitute a threat.

What helps you identify a profile of the offender?

We won't try to provide a psychological profile of a writer or describe a writer's intent. I won't, for example, be able to determine whether a person is lying or how likely they are to carry out a threat. We do sometimes work alongside forensic psychologists who may come to judgements about these issues.

We can describe a sociolinguistic profile, however. Language varies by social group. One example would be the substantial academic literature on language differences between men and women. We use this literature but turn it on its head. Rather than taking a range of writings by men and a range by women and describing the differences between them, we take a single text and ask, from our knowledge of the generalities, which group it belongs to. This reversal of normal scientific method is why it's so easy to get it wrong and why this sort of work should remain inadmissible as evidence.

Also different social groups sometimes use different dialects. In one case the phrase "bad-minded men" was used. This was an unusual phrase to me but is fairly common in the Jamaican-Caribbean community, so use of the phrase seems to suggest membership of, or contact with, that group.

Writers fall into habits and patterns not only of language but also of writing behaviours. Sometimes you'll notice the letters or emails in a case are only sent on Sundays, for example, and this may tell you something about the writer or fit with something else the police know.

How do computer analyses work? What do they pick up that you cannot?

We can also use statistical and computer analyses, which can pick up patterns that may not be obvious to the eye. By looking at a variety of language statistics we can inform or challenge our more qualitative judgements.

This is more developed in comparative authorship analyses where we know, for example, that the rate of use of functional words (short words like 'of', 'and', 'the' or 'but') can be used to distinguish between authors. This works well with longer texts and we've been developing alternative techniques for short texts such as SMS text messages and online chat.

We also use word-frequency distributions, which are statistical measures of the richness and diversity of a person's vocabulary. These are useful in comparitative cases and there's some research suggesting that these measures can also be useful in profiling cases, for example to identity older writers.

The computational and statistical work in the profiling area is at the leading edge. We've got postgraduate students looking at non-native writers in English and identifying their first language through statistical analyses of their language.