For the first time, a joint laboratory on artificial intelligence from the Chinese mainland has climbed to the top of the leaderboard of Stanford University’s reading comprehension task.

The joint laboratory of Harbin Institute of Technology and iFLYTEK Research now sits on top of Stanford Question Answering Dataset (SQuAD) leaderboard, followed by Microsoft Research Asia in second place and IBM Research in tenth place. Other Chinese institutes are also on the board, including Eigen Technology and Zhejiang University and Tsinghua University.

(Snapshot of SQuAD website)

SQuAD is a new reading comprehension dataset, consisting of questions posed by crowdworkers on a set of Wikipedia articles, where the answer to every question is a segment of text, or span, from the corresponding reading passage.

With over 100,000 question-answer pairs on some 500 articles, SQuAD is significantly larger than previous reading comprehension datasets, according to its website.

The Chinese joint laboratory scored an accuracy rate of 77.845 on exact matching, while runner-up Microsoft got 77.688 percent correct.

“It is relatively easy for machines to conduct simple deduction based on massive memorization. But it is harder to comprehend and deduct in a precise way. Almost all AI research teams are working on similar things,” Wang Shijin, deputy director of the champion joint laboratory, told Thepaper.cn.

Since May 2015, the joint laboratory preformed machine reading comprehension tasks. Specifically, the lab aims to make the machine read at the sixth-grade level, while one of iFLYTEK’s grander goals is to understand China's college entrance examination, as it hopes to make machine scores high enough for top universities.

In June, a Chengdu-designed artificial intelligence program called AI-MATHS scored 100 out of 150 on the national test paper on math in just 10 minutes.