The Chinese tech giant Baidu has outperformed Google and Microsoft in an artificial intelligence competition designed to test the machine’s ability to understand human language, with Baidu, often referred to as Google China, achieving the highest score in assessing the understanding of the general English sentence (GLUE), which is The widely accepted criterion for language understanding for AI.
Ernie's corporate model (Enhancing Representation Through Knowledge Integration) became the first to score more than 90 points in the test, topping the list of leaderboards dominated by US technology companies and universities, and this work makes the Ernie model one of only ten artificial intelligence systems that outperform average Human scores of 87.1 on the GLUE scale.
GLUE consists of nine different tests, with each test measuring a different task important for understanding the language, such as distinguishing between entity names in a sentence and discrimination in which context the pronoun is used when there are many potential candidates, and the average human usually scores about 87 points on the GLUE out of 100 Potential point.
The Ernie model used a method similar to the Google Bert language model, which attempted to understand the natural language of AI when it was created last year, and both the Bert and Ernie model interpret the meaning by examining the words that appear before and after each word in the sentence to fully establish the context.
By first developing the Ernie model using Chinese and then using English words, Baidu researchers realized that they made the algorithm stronger in understanding the English language, and the Chinese company actually uses the model to improve results for its search engine and make its assistant for artificial intelligence named Xiao Du more accurate .
`` When we first started this work we were especially thinking about specific characteristics of the Chinese language, but we quickly discovered that the model was applicable even further, '' Hao Tian, Baidu's chief research engineer, told MIT Technology Review which first reported the research.
It is reported that the Chinese company intends to present a research paper in which it explains in detail how the Ernie model was trained in language testing at the Association for the Advancement of Artificial Intelligence next year. The paper also shows that the model now leads all other models, and that the algorithm capable of better understanding the English language is Those designed to learn Chinese.
0 Comments