@misc{indiciae91606ae45988,
  title = {Latent Positional Information is in the Self-Attention Variance of Transformer Language Models Without Positional Embeddings},
  author = {Ta-Chung Chi and Ting-Han Fan and Li-Wei Chen and Alexander I. Rudnicky and Peter J. Ramadge},
  year = {2023},
  url = {https://arxiv.org/abs/2305.13571},
  note = {Source identifier: 2305.13571}
}