http://www.mikeash.com/?page=pyblog/friday-qa-2018-04-27-generating-text-with-markov-chains-in-swift.html#commentsmikeash.com Recent CommentsSat, 06 Jun 2026 20:00:41 GMTPyRSS2Gen-1.0.0http://blogs.law.harvard.edu/tech/rsshttp://www.mikeash.com/?page=pyblog/friday-qa-2018-04-27-generating-text-with-markov-chains-in-swift.html#commentsNLTokenizer in 12 beta/10.14 beta looks to be really useful for something like this. <br /> <br /><a href="https://developer.apple.com/documentation/naturallanguage/nltokenizer?changes=latest_minor">https://developer.apple.com/documentation/naturallanguage/nltokenizer?changes=latest_minor</a>53695090144d50676dab9922aa4f9b9eWed, 13 Jun 2018 17:11:02 GMThttp://www.mikeash.com/?page=pyblog/friday-qa-2018-04-27-generating-text-with-markov-chains-in-swift.html#commentsCould a method like this also be used for forensics - determining if a given body of text was written by a suspect? <br /> <br />For example, if you had a large body of writing from User_123 and used it to populate the probabilities of a Markov chain, you would expect that another large body of writing from the same user would generate similar probabilities. So given a body of writing from an unknown user, you could compare the probabilities and determine whether or not User_123 wrote both. Although I imagine this approach would have quite large error bars - and would be more error prone as you dealt with smaller and smaller bodies of writing. <br /> <br />I'm also guessing the accuracy of the above approach would be affected by the topic covered by the body of writing - for instance I'm guessing the Markov chain generated from your blog would be very different from one generated from your text messages to family members - probably fewer references to ARC and APIs. <br /> <br />In any case - very interesting write up!5646d1bc15083be4be8906594e4854adMon, 30 Apr 2018 19:04:39 GMT