For decades, linguists seeking to sort out the history of language sifted through words and grammars by hand in search of clues to tongues’ shared pasts. SFI Professor Tanmoy Bhattacharya and External Professor Peter Stadler are taking the first steps toward an improved approach during a March working group at SFI on the methods and mathematics of historical linguistics.
“The current standard in this field is you ask which words are cognate [have the same origin],” and then consider the words’ grammatical and semantic contexts to infer whether two languages are related, Bhattacharya says. The better approach is to infer the probability that words from different languages are cognate, then “propagate the uncertainty” to determine the probability that entire languages are related, he says.
It’s a problem a number of SFI scientists have worked on, often taking their inspiration from biological phylogenetics – the process of inferring evolutionary history from the genetic similarity of modern-day plants and animals. It’s a useful starting point, Bhattacharya says, but language is doing something different from biology. While the basic processes of biology are fixed in time – “a G is a G,” he says, referring to the DNA nucleotide guanine – the relationships between phonemes, pronunciation, and word meanings all drift.
With so many moving parts at work in language evolution, the standard mathematical and computational tools researchers use to search for common ancestors in biology won’t work.
“We have to have a different algorithm for fitting the data,” Bhattacharya says.
Equally important is finding a way to be more objective about what patterns in the data are real. To be sure, linguists have discovered real patterns that can be useful checks on a computer model, but “what we need is an automatic way to find these things.” That, he says, “is better done by machines than by humans.”
The working group brings together physicists, biologists, anthropologists, and others who have spent years working on the problem. The top question, Bhattacharya says, is: “How do we make real progress beyond what we’ve done individually?”