A new work published in the Proceedings of the National Academy of Science details how researchers developed an automated system that accurately reconstructs ancient languages that evolved into modern language.
The system was applied to 637 languages spoken primarily in Southeast Asia. Typically, the parent languages are manually reconstructed using a process called the comparative method. The automated system is proven effective, as it has reconstructed over 85 percent of the language within one character of the manual reconstruction method.
Since the 18th century, reconstructing languages has been difficult for linguists, and various methods were developed to resolve this problem. Using the comparative method, information is drawn from the sounds and words found in modern language and used to hypothesize.
"What excites me about this system is that it takes so many of the great ideas that linguists have had about historical reconstruction, and it automates them at a new scale: more data, more words, more languages, but less time," said Dan Klein, an associate professor at the University of California at Berkeley in a university statement.
When using the system, it's possible to complete a large-scale reconstruction of ancient languages directly from modern language words, using algorithms and probabilistic models of sound change. The ability to do this makes it possible to understand the factors that are involved in language change. This lead to the more in-depth investigation of a linguistic theory called the "functional load," regularity in sound changes. This hypothesis was initially introduced in 1955 by Andre Martinet and "asserts that sounds that play a more important role in distinguishing words are less likely to change over time," the report says.
In 1967, this hypothesis was tested by defining a statistic to determine the amount of information lost when sound change is lost in a language. Little evidence was found, as the theory was tested among only four languages. Using the new system, researchers collected the sound change data from the 637 reconstructed languages and obtained the information required for a better evaluation of the hypothesis. The new system goes far beyond the capabilities of the manual methods and can also be used as a tool to investigate the structure and dynamics of languages.
"It would take hundreds of lifetimes to pore over all those languages, cross-referencing all the different changes that happened across such an expanse of space — and of time. But this is where computers shine," said Klein.
