{
  "id": "neural-linguistic-steganography",
  "title": "Neural Linguistic Steganography",
  "authors": [
    "Zachary M. Ziegler",
    "Yuntian Deng",
    "Alexander M. Rush"
  ],
  "author_parts": [
    {
      "given": "Zachary M.",
      "family": "Ziegler"
    },
    {
      "given": "Yuntian",
      "family": "Deng"
    },
    {
      "given": "Alexander M.",
      "family": "Rush"
    }
  ],
  "equal_contribution": [
    "Zachary M. Ziegler",
    "Yuntian Deng"
  ],
  "year": 2019,
  "venue_date": "2019-11",
  "venue": "Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP)",
  "venue_short": "EMNLP-IJCNLP 2019",
  "publication_type": "conference",
  "publisher": "Association for Computational Linguistics",
  "pages": "1210–1215",
  "doi": "10.18653/v1/D19-1115",
  "arxiv_id": "1909.01496",
  "project_url": "https://steganography.live/",
  "relation": "foundation",
  "is_original_project": true,
  "categories": [
    "Original project",
    "Arithmetic coding",
    "Language-model steganography"
  ],
  "summary": "Combines arithmetic coding with large pretrained language models to encode secret information into generated text while closely matching the model's distribution. GPT-2 experiments evaluate information capacity, statistical divergence, and human judgments of contextual continuations, and demonstrate encoding natural-language messages into unrelated cover text.",
  "relationship": "The original paper associated with this research site and its STEGASURAS implementation. The published paper links steganography.live and the official repository.",
  "limitation": "The statistical experiments measure divergence from the language model, not an inaccessible true human-language distribution. Truncation and temperature choices create a practical trade-off between human plausibility and distribution matching.",
  "sources": [
    {
      "url": "https://aclanthology.org/D19-1115/",
      "evidence": "Publication title, venue, November 2019, DOI and pages."
    },
    {
      "url": "https://aclanthology.org/D19-1115.pdf",
      "evidence": "Title page establishes full author names and equal contribution; sections 3–4 support method, evaluations and limitations; page 2 links site and code."
    },
    {
      "url": "https://arxiv.org/abs/1909.01496",
      "evidence": "Only listed version is v1, September 3, 2019."
    },
    {
      "url": "https://github.com/harvardnlp/NeuralSteganography",
      "evidence": "Official implementation identifies paper and links steganography.live; June 2025 compatibility update is separately credited."
    }
  ],
  "arxiv_url": "https://arxiv.org/abs/1909.01496",
  "earliest_date": "2019-09-03",
  "earliest_date_label": "First arXiv submission",
  "url": "https://aclanthology.org/D19-1115/",
  "pdf": "https://aclanthology.org/D19-1115.pdf",
  "code": "https://github.com/harvardnlp/NeuralSteganography",
  "category": "original",
  "contribution": "Combines arithmetic coding with large pretrained language models to encode secret information into generated text while closely matching the model's distribution. GPT-2 experiments evaluate information capacity, statistical divergence, and human judgments of contextual continuations, and demonstrate encoding natural-language messages into unrelated cover text.",
  "canonical_url": "https://steganography.live/reading-list/papers/neural-linguistic-steganography/"
}