mutiann / neural-lexicon-reader Goto Github PK

Thanks a lot for publishing your wonderful work. In you paper, "In our experiments,
we use pretrained XLM-R as the knowledge encoder, using its final-layer embeddings as keys to represent semantic
information. They are then concatenated with the first-layer
embeddings as values to inject shallow character identities." However, in your code, you take the first-layer
embeddings as keys.