Students Win International Computational Linguistics Competition

Roxana Girju, Beckman Institute faculty member in the Artificial Intelligence group, recently led a team of University of Illinois and Beckman Institute students to victory in an international competition sponsored by the Association for Computational Linguistics, a society for people working on problems involving natural language and computation.

Team members included Alla Rozovskaya, Andrew Fister, Suma Bhat, Brant Chee, and Brandon Beamer.

The Illinois students built a "semantic parser" that ranked first in the competition on Semantic Relations between Nominals (Task 4) at the Semantic Evaluation (SemEval-4) series. Semantic parsing between nominals (nouns) is the task of identifying the underlying meaning encoded by two nouns in a sentence.

Fourteen teams took part in the competition, including university or company teams from Europe, Australia, Japan, and the United States. The competition is held once every three years and this is the first time a team from the University of Illinois participated.

Each team received a list of relation classes from several different classification schemes, along with trial and training data. From there they built "learning classifiers." The classifiers were then applied to label new, unseen instances.

"Recognizing with high accuracy such relations is difficult due to the inherent ambiguity of the English language," said Girju. Girju also noted that semantic relations are important since they can serve as building blocks for high-level applications such as search engines, question answering, and textual inference.

Girju said the semantic parser task is viewed not only as a competition, but as an evaluation task that allows researchers to compare their approaches and to make important advances natural language understanding.

"The aim of computational semantics is to identify techniques for semantic representations of natural language text -- representations that can be used to perform inference," Girju said. "The inherent ambiguity of natural languages makes it a challenging topic, but it is mandatory for any advanced natural language understanding system."

Girju said semantic parsers, such as the award-winning student design, are going to bring more insight into various research issues such as artificial intelligence, linguistics, and cognitive science.