{
  "annotation_id": "5a3388ca-f9a9-4ea2-81d4-9f5c2d2e737e",
  "date_created": "2026-02-17T05:53:23.284000Z",
  "date_modified": "2026-02-17T05:53:23.284000Z",
  "file_hash": "6f849f5eab9821bd63b6a2002744cfdd229118f5db13abd1342afce820cd01da",
  "private": false,
  "record": {
    "abstract": "Compositionality is a cognitive mechanism that allows humans to systematically combine known concepts in novel ways. This study demonstrates how artificial neural agents acquire and utilize compositional generalization to describe previously unseen images. Our method, termed \"Composition through Decomposition\", involves two sequential training steps. In the \u0027Decompose\u0027 step, the agents learn to decompose an image into basic concepts using a codebook acquired during interaction in a multi-target coordination game. Subsequently, in the \u0027Compose\u0027 step, the agents employ this codebook to describe novel images by composing basic concepts into complex phrases. Remarkably, we observe cases where generalization in the `Compose\u0027 step is achieved zero-shot, without the need for additional training.",
    "arxiv_id": "2601.10169",
    "authors": [
      "Boaz Carmeli",
      "Ron Meir",
      "Yonatan Belinkov"
    ],
    "categories": [
      "cs.AI"
    ],
    "license": "http://creativecommons.org/licenses/by/4.0/",
    "title": "CtD: Composition through Decomposition in Emergent Communication",
    "url": "https://arxiv.org/abs/2601.10169",
    "version": "v1"
  },
  "schema_id": "dorsal/arxiv",
  "source": {
    "execution_id": "d6e22643-8909-4ba0-aa3a-28983f2733c9",
    "id": "arXiv Dataset",
    "type": "Model",
    "variant": "snapshot-2026-01-17",
    "version": "0.1.0"
  },
  "user_id": 1000002
}