Researchers at University College London in the U.K. have developed a chess algorithm that evaluates the quality of chess moves by analyzing the reactions of expert commentators.
The researchers analyzed the text of 2,700 chess match commentaries available online, removing commentary that did not relate to high-quality moves, and examples that were too ambiguous.
They used a special type of recurrent neural network, and a mathematical technique called word embeddings that connects words on the basis of their meanings.
The team found that the SentiMATE algorithm was able to work out some of the basic tenets of chess, as well as several key strategies, such as forking and castling.
While the algorithm failed to beat some conventional chess bots consistently, it demonstrates the potential of using language to help determine how to play the game well, using less practice data and less computer power than other chess programs require.
From Technology Review
View Full Article
Abstracts Copyright © 2019 SmithBucklin, Washington, DC, USA