@inproceedings{v-s-tef-anik-etal-2026-language,
    title = "Language Models Learn Universal Representations of Numbers and Here{'}s Why You Should Care",
    author = "{\textbackslash}v{\{}S{\}}tef{\textbackslash}'anik, Michal  and
      Mickus, Timothee  and
      Kadl{\textbackslash}v{\{}c{\}}{\{}{\textbackslash}'{\textbackslash}i{\}}k, Marek  and
      H{\{}{\textbackslash}o{\}}jer, Bertram  and
      Spiegel, Michal  and
      V{\textbackslash}'azquez, Ra{\textbackslash}'ul  and
      Sinha, Aman  and
      Kucha{\textbackslash}v{\{}r{\}}, Josef  and
      Mondorf, Philipp  and
      Stenetorp, Pontus",
    editor = "Liakata, Maria  and
      Moreira, Viviane P.  and
      Zhang, Jiajun  and
      Jurgens, David",
    booktitle = "Proceedings of the 64th Annual Meeting of the {A}ssociation for {C}omputational {L}inguistics (Volume 1: Long Papers)",
    month = jul,
    year = "2026",
    address = "San Diego, California, United States",
    publisher = "Association for Computational Linguistics",
    url = "https://preview.aclanthology.org/ingest-acl/2026.acl-long.1415/",
    pages = "30663--30681",
    ISBN = "979-8-89176-390-6",
    abstract = "Prior work has shown that large language models (LLMs) often converge to accurate input embedding for numbers, based on sinusoidal representations.In this work, we demonstrate that these representations are in fact strikingly systematic, to the point of being almost perfectly universal: different LLM families develop equivalent sinusoidal structures, and number representations are broadly interchangeable in a large swathe of experimental setups.We show that properly factoring in this characteristic is crucial when it comes to assessing how accurately LLMs encode numeric and other ordinal information, and that mechanistically enhancing this sinusoidality can also lead to reductions of LLMs' arithmetic errors."
}