Markov & statistical models unlock ancient script

Agree about the bad title; but it does resolve an important question, which was whether the script was a language at all. Establishing that it has the regularities of human language was a necessary first step.

Here's another interesting example. The Luwian script's directionality couldn't be established. It was not certain whether it was up-down, up-down or up-down, down-up or various possibilities. This paper established using Expectation-Maximization the correct order of reading for the script:

Discovering the linear writing order of a two-dimensional ancient hieroglyphic script http://www.csie.ntu.edu.tw/~sdlin/publications.html

Luwian script: http://en.wikipedia.org/wiki/Hieroglyphic_Luwian

> "One of the main purposes of our paper is to introduce Markov models, and statistical models in general, as computational tools for investigating ancient scripts," Adhikari said.

This is bogus. (Hidden) Markov Models and other generative probabilistic models have been used in all kinds of decipherment for years - see the latter half of http://www.isi.edu/natural-language/people/naacl09-print-1x2... for an overview.

Modern statistical machine translation works by training on large amounts of already translated data, called parallel corpora. Monolingual data is helpful and can help you fix grammar, choose words that are more likely, but it won't work without that parallel corpora. Basically, you need a Rosetta stone, and if you had that, the humans could slowly deduce all sorts of things from only a small amount of data that would probably be useless for an MT system. Actually, scanning, or OCR, is a similar process--you train first on text that already has been properly transcribed, then you have a model that can be used to do OCR on new data.