A General Method for Transferring Explicit Knowledge into Language Model Pretraining

Figure 4

Performance of K-XLNet models injected with different triplet (knowledge) amounts.