Baidu Tops AI Competition Designed to Test How Well a Machine Can Understand Human Language
Mon, April 19, 2021

Baidu Tops AI Competition Designed to Test How Well a Machine Can Understand Human Language

Baidu's Enhanced Representation through the kNowledge IntEgration model (ERNIE) recorded the highest score of 90.1 on GLUE benchmark / Credits: PixieMe via Shutterstock

 

Chinese tech companies strongly competing with Western tech companies in advancing their technologies, especially in using artificial intelligence. The recent AI competition is about which company can do better in the General Language Understanding Evaluation (GLUE) benchmark, a collection of resources for training, evaluating, and analyzing natural language understanding systems. 

Recently, Baidu announced that its Enhanced Representation through the kNowledge IntEgration model (ERNIE) has beaten Google and Microsoft in an AI competition designed to test how well a machine can understand human language. ERNIE recorded the highest score of 90.1 on GLUE benchmark, widely considered to be the benchmark for AI language understanding. The model was first developed to understand the Chinese language but researchers discovered that it can understand English as well. 

ERNIE is the first model to become the first to score above 90 on the test, topping a leaderboard dominated by US tech firms and universities. According to the South China Morning Post, a Hong Kong English-language newspaper, since ERNIE can perform better in both English and Chinese, Baidu has adopted it for real-world applications by using the AI model to deliver better search results. This also makes the model one of only 10 AI systems to surpass the average human score of 87.1 on the GLUE benchmark.

"When we first started this work, 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, said. 

According to The Independent, a British news publication, ERNIE used a similar method to Google’s Bidirectional Encoder Representations from Transformers match (BERT), which transformed natural-language understanding for AI when it was created last year. Both ERNIE and BERT can interpret the meaning as they can examine the words that appear both before and after a word in a sentence to establish context.