Beijing, China-based search engine giant Baidu has overtaken Google and Microsoft in an AI (artificial intelligence) competition that was designed to see how well a machine could understand human language.
Baidu, which is referred to as China's Google, scored the highest score in GLUE (General Language Understanding Evaluation), a widely-accepted benchmark for AI language understanding system.
The firm’s Ernie (Enhanced Representation through kNowledge IntEgration) model became the first to score above 90 points, also surpassing the average human score of 87 points on GLUE, out of a possible 100. Ernie, inspired by US technology giant Google’s Bert (Bidirectional Encoder Representations from Transformers) model, transformed natural-language understanding for AI.
Both the models, named after the Sesame Street characters Bert and Ernie, interpret meaning by examining the words that appear both before and after a word in a sentence to fully establish context. Google’s model hides 15% of the words in each sequence and then predicts them based on the context.
Baidu first used Chinese language and then tested the algorithm in English, and the researchers found that it performed better in the latter language.
“When we first started this, we were thinking specifically about certain characteristics of the Chinese language. But we quickly discovered that it was applicable beyond that,” Hao Tian, chief architect of Baidu Research, told MIT Technology Review.
GLUE consists of nine different tests for things like picking out the names of people and organisations in a sentence, and figuring out where a pronoun like ‘it’ should be used when there are numerous potential candidates.