Search papers, labs, and topics across Lattice.
Aarhus University
1
0
3
Multilingual embeddings that seem to encode language proficiency in one dataset completely fail to generalize to others, suggesting they're picking up dataset-specific artifacts, not true proficiency.
