Investigating Cross-Lingual Alignment Methods for Contextualized Embeddings with Token-Level Evaluation