BibTeX

@inproceedings{nikolaev23:_repres,
  abstract = {Variants of the BERT architecture specialised for producing full-sentence representations often achieve better performance on downstream tasks than sentence embeddings extracted from vanilla BERT. However, there is still little understanding of what properties of inputs determine the properties of such representations. In this study, we construct several sets of sentences with pre-defined lexical and syntactic structures and show that SOTA sentence transformers have a strong nominal-participant-set bias: cosine similarities between pairs of sentences are more strongly determined by the overlap in the set of their noun participants than by having the same predicates, lengthy nominal modifiers, or adjuncts. At the same time, the precise syntactic-thematic functions of the participants are largely irrelevant.},
  added-at = {2023-01-23T10:54:40.000+0100},
  address = {Dubrovnik, Croatia},
  author = {Nikolaev, Dmitry and Pad{ó}, Sebastian},
  biburl = {https://puma.ub.uni-stuttgart.de/bibtex/2c0170446a363bca46ae6b2da016b151d/sp},
  booktitle = {Proceedings of EACL},
  interhash = {2dd0ba5490c2846139e7654557e7eb14},
  intrahash = {c0170446a363bca46ae6b2da016b151d},
  keywords = {conference myown},
  timestamp = {2023-05-02T15:21:40.000+0200},
  title = {Representation biases in sentence transformers},
  url = {https://aclanthology.org/2023.eacl-main.268},
  year = 2023
}