Refining a Computer's Understanding of Language

University of Maryland researcher Shomir Wilson has developed methods for computers to determine when people are describing words and phrases.

Further research on this topic will make it possible for computers to understand the information conveyed by metalanguage--statements describing words and phrases--leading to improvements in technologies such as conversational computer systems and automated dictionary building tools.

People frequently use metalanguage to introduce new words, clarify meaning and correct mistakes. But previously, the clues in vocabulary and grammar that identify metalanguage were unknown.

To study metalanguage, Wilson collaborated with other computer scientists at the National University of Singapore and Macquarie University in Australia. Many samples of metalanguage were gathered from existing language resources and then examined for recurring patterns. Using these patterns, computer algorithms called classifiers were trained to automatically detect metalanguage in new text. A small set of clues in vocabulary and grammar were very common and accurate as predictors.

In the future, these methods will make it easier to interact with computers in the same natural language that people use when interacting with each other.