How is a kidnapper's text message similar to a jellyfish? Both are tiny points in a sea of data that scientists can use to draw conclusions about a bigger picture -- be it identifying a serial rapist or measuring linkurl:biodiversity,;http://www.the-scientist.com/article/display/53881/ according to linkurl:Andrew Price,;http://www.bio.warwick.ac.uk/res/frame.asp?
How is a kidnapper's text message similar to a jellyfish? Both are tiny points in a sea of data that scientists can use to draw conclusions about a bigger picture -- be it identifying a serial rapist or measuring linkurl:biodiversity,;http://www.the-scientist.com/article/display/53881/ according to linkurl:Andrew Price,;http://www.bio.warwick.ac.uk/res/frame.asp?ID=36 a marine ecologist who is working with forensic scientists to apply taxonomic techniques to crime assessment.
Ecologists routinely track biodiversity in different marine environments, and those same techniques can be used to assess similarities and differences in details relating to a linkurl:crime,;http://www.the-scientist.com/news/display/53864/ in order to determine, for example, if two assaults were committed by the same person.
The connection is obvious, said linkurl:Tim Grant,;http://www.forensiclinguistics.net/cfl_staff.html deputy director of the Center for Forensic Linguistics at Aston University in Birmingham, UK, who sought out Price's expertise. "What we're all interested in doing is measuring similarity."
Grant and his collaborator, forensic psychologist linkurl:Jessica Woodhams,;http://resources.glos.ac.uk/faculties/ehs/sciences/staff/jwoodhams.cfm first connected with Price about three years back. Grant and Woodhams, then both at the University of Leicester, were trying to link crimes perpetrated by the same offender. They sent a query out on an international E-mail list in search of methods that might help analyze a database of details from police investigations of sexual offenders.
Price, conveniently based just 30 miles down the road at the University of Warwick, saw their note and thought, "Ahh, this could be interesting." A few years back, he had developed an index of taxonomic similarity for use in assessing marine biodiversity. Most biodiversity measures track whether certain species are present or absent in an environment. That means that an ocean region containing, say, a species of crab, whale, snail, coral, and jellyfish would be ranked as having a similar level of biodiversity as a region containing five different crab species, despite the fact that the former has much greater taxonomic spread.
"Virtually all the measures that people have used have focused on species as the end-all, be-all," Price said. His technique, on the other hand, aimed to incorporate additional taxonomic information by classifying species by similarities. "If two sites didn't share the same species, but shared the same genus, then that would be worked into the analysis," he explained. "It's making use of information up the hierarchy, and it's giving a weighting at each successive level."
Forensics has a similar problem: Investigators try to identify serial offenders by pegging similarities in specific behaviors, such as luring subjects with a particular brand of alcohol, or wearing a particular item of clothing. But violent crime victims might not remember such details, or the police may not think to ask.
By hierarchically ranking the information in their sex offenders database, the trio found, they were able to linkurl:use it more effectively and systematically;http://www3.interscience.wiley.com/journal/114273993/abstract in their analysis. "What it's essentially doing is tapping into less precise information," said Price. "With our measure, you could say, 'Well, ok, Chardonnay wasn't offered [to the victim], but was there any alcohol?'"
Indeed, even when they removed 50% of the information on specific behaviors in their data set, they were still able to show a significant difference between serial offenses and random ones.
In the course of his collaboration with Price, Grant said, he's "learned a fair bit about how taxonomies are formed." He and the marine ecologist are now extending the approach to identifying perpetrators of other types of crimes, such as theft, terrorism, and bomb threats.
Grant said the technique will prove especially useful in his area of special interest -- forensic analysis of text messaging. In a recent murder case, an analysis of text messages sent from the phone of a Yorkshire teenager named Jenny Nicholl suggested that the texts were sent not by her, but by a man suspected of killing her; she generally typed out "I am" and "myself" in her text messages, but after she disappeared, the style shifted to "im" and "meself."
Since 2006 his group has been collecting a database of text messages contributed by members of the public -- you can linkurl:add to the database here;http://www.forensiclinguistics.net/texting -- to analyze variation and consistency in an individual's texting style. The idea, he said, is, "Can you tell anything about the person who wrote this? Is it a 13-year-old girl or a 50-year-old man?"
"What it does is focus on the method rather than the [forensic scientist's] expertise," Grant added. "Because it's a describable, repeatable method, it's closer to a forensic science."
linkurl:Lawrence Solan,;http://www.brooklaw.edu/faculty/profile/?page=155 a law professor with a background in linguistics at Brooklyn Law School, agreed. "I think it's a breath of fresh air that people working in forensic identification sciences are looking at methodologies used in core basic sciences and seeing what they can learn from them," he said. Still, Solan said, ferreting out differences in forensic evidence is just as important as identifying similarities. For text analysis, he noted, "the real test is whether [the technique] can take something that's reasonably similar, but not by the same author, and tell us that."
For Price, the collaboration has had an unexpected benefit: testing his analysis on a known data set will help refine the technique for use in ecology. "What's interesting in marine ecology is that you can't do a real experiment," he said. Determining the differences between two different sites is a qualitative endeavor -- "they're kind of partially different, but there's no real answer." Essentially, the body of information on known repeat and one-off sexual offenders allows him to test and refine the method experimentally, he said, adding, "We just put in a grant submission to see how it benefits both directions."
September 26, 2008
"Ahh, this could be interesting," he said. And indeed it was! Thanks for an excellent story.
September 27, 2008
It will be really interesting to see how this methodology develops, particularly given the point that its focus is on a describable, repeatable method rather than the forensic scientist's expertise. \n\nCarole Chaski PhD. director of the Institute for Linguistic Evidence notes that the primary difference between forensic linguistics and non-linguist methods is the scientific approach. According to Dr. Chaski, "In forensic linguistics, the scientific method requires hypothesis testing and a litigation-independent testing of the method for its accuracy. Further, these tests are performed with robust controls regarding data quantity, data sources and analytical objectivity." ( www.all-about-forensic-science.com
September 30, 2008
Thank you to annonymous who gave the cool web address for forensic science.