Learning vocabulary like a human

A team of researchers at Carnegie Mellon University has developed a computer system that is designed to learn the English language as a human would – cumulatively, over time. Called NELL (for “Never-Ending Language Learning system”), the system “reads” millions of texts from the web to accumulate vocabulary items and notice semantic patterns among them (e.g., whether the item is a person, place, plant). The system is able to not only add to its knowledge and categorize the facts it learns, but can review what it has learned, and revise that knowledge if necessary. In other words, it keeps “learning” from what it has learned. It is this feature that makes the system seem to have a more human-like way of learning language. So far NELL is learning only English, but the system has applications for all languages.

This topic would be an interesting way to combine a focus on technology with a discussion of vocabulary learning for technical students who might not be aware of the linguistic processes involved in learning a language. If they see how both their brain and a computer is programmed for learning a language, it could make them more aware of their own language-learning skills.

This is also a way for teachers to help students develop good vocabulary-building skills (and relate them to reading skills). NELL’s focus on semantic fields and categories (part of how NELL learns vocabulary items) is similar to a good reader’s ability to make a connection among words and concepts – a skill students should be encouraged to develop. This can be referred to each time students discuss a text they have read, and can also help students focus on how they can develop effective vocabulary-building skills.

Two articles about NELL:

From ScienceBlog (also has an interview with one of NELL’s developers, useful for discussing the technical aspect of the project)

http://scienceblogs.com/universe/2010/10/08/read-the-internet-speak-englis/

From The New York Times http://www.nytimes.com/2010/10/05/science/05compute.html?pagewanted=1&_r=1&ref=science