内容改写 However these works only indirectly measure the effectiveness of the reasoning path by evaluating downstream tasks Becker et al 2021 automatically evaluates the knowledge path through the simila

Nonetheless, the effectiveness of the reasoning path is only assessed indirectly in these studies, as they evaluate downstream tasks. In contrast, Becker and his colleagues (2021) measure the knowledge path by comparing it to the implicit knowledge in the dataset. However, their method requires an annotated golden path to work properly.