Robert Östling
Robert Östling

In order to understand Robert Östling’s contributions to the field, it is necessary to know what is meant by computational linguistics, the field in which he wrote his doctoral thesis at the Department of Linguistics. Without thinking about it, most of us make use of computational linguistics every day, for example, when we search for information about a topic on Google, when we run the spellcheck program, or when we want to translate words or full sentences into a foreign language.

“To put it very simply, computational linguistics is about making computers process and convert human languages,” says Robert Östling.  It is often about finding out if the person writing has a positive or negative attitude to what he or she is writing. It is also about creating computer programs that can learn to use a text that has already been translated to translate a new text that it has not seen before.

“This is not easy, since computers think with ones and zeros without understanding the content the way we humans do. That is why you often get weird translations,” explains Robert Östling.

Better alignment methods

In his own research, Robert Östling seeks to improve the methods for aligning words and phrases in different languages. Furthermore, he has applied these methods in order to learn something about the structure of the languages. In his thesis, he studies Swedish Sign Language and 1,001 different languages around the world.
“The reason I have studied sign language is that my department has collected and transcribed conversations in sign language. I have then linked them to a Swedish translation in order to divide the signs into word classes using the Swedish text.”

When it comes to foreign languages, Robert Östling has used the Bible’s New Testament, which has been translated into thousands of languages. Many of these translations are available on the Internet. Linking the words in 1,001 languages at the same time is one of the novelties in the thesis; this is usually done between two languages at a time.

“I wanted to explore how to do it with many different languages. I have been able to transfer information from the Bible to all 1,001 languages that we have translations for. Using the Bible texts to examine the structure of different languages, we can, in addition to dividing the words into word classes, find patterns in the languages’ arrangement of clause elements; for example, if the object is usually placed after the verb, as in Swedish, or the other way around, as in Japanese,” says Robert Östling.

New algorithms

The method used in Robert Östling’s work is based on Bayesian models that utilise existing knowledge about a topic and then improve it on the basis of observations.

“This is an umbrella term for a type of statistical models that have become increasingly common in fields such as computational linguistics,” says Robert Östling.

In his thesis, he has also developed new algorithms that make it possible to better utilise the information in texts that have been translated into hundreds or thousands of languages.

“Those who can make the most practical use of my work are linguistics. Thus far, they have mostly used books and carried out manual analyses. Now they can more easily retrieve the information automatically in the computer,” says Robert Östling.