pith:OUDWHWEU
How Language Models Process Negation
Language models process negation by constructing representations of negative phrases more than by suppressing positives.
arxiv:2605.03052 v2 · 2026-05-04 · cs.CL
Add to your LaTeX paper
\usepackage{pith}
\pithnumber{OUDWHWEUIP57XMCUMOAYY3MKL7}
Prints a linked badge after your title and injects PDF metadata. Compiles on arXiv. Learn more · Embed verified badge
Record completeness
Claims
We apply a range of observational and causal interpretability techniques on Mistral-7B and Llama-3.1-8B to show that models implement both mechanisms, with the 'constructive' mechanism being more prominent.
That the observational and causal interpretability techniques (such as attention ablation and activation analysis) accurately isolate and identify the specific mechanisms responsible for negation processing without confounding effects from other model components.
LLMs implement both attention-based suppression and constructive representations for negation, with construction dominant, despite poor accuracy from late-layer attention shortcuts.
Formal links
Receipt and verification
| First computed | 2026-06-02T01:03:16.146064Z |
|---|---|
| Builder | pith-number-builder-2026-05-17-v1 |
| Signature | Pith Ed25519
(pith-v1-2026-05) · public key |
| Schema | pith-number/v1.0 |
Canonical hash
750763d89443fbfbb05463818c6d8a5fc7b6f0892d3b00df9f89237a8b4050c9
Aliases
· · · · ·Agent API
Verify this Pith Number yourself
curl -sH 'Accept: application/ld+json' https://pith.science/pith/OUDWHWEUIP57XMCUMOAYY3MKL7 \
| jq -c '.canonical_record' \
| python3 -c "import sys,json,hashlib; b=json.dumps(json.loads(sys.stdin.read()), sort_keys=True, separators=(',',':'), ensure_ascii=False).encode(); print(hashlib.sha256(b).hexdigest())"
# expect: 750763d89443fbfbb05463818c6d8a5fc7b6f0892d3b00df9f89237a8b4050c9
Canonical record JSON
{
"metadata": {
"abstract_canon_sha256": "3f7d4817f55118373e90a8f549904e01dab05f60d99547f0ba79e90118fd7d0c",
"cross_cats_sorted": [],
"license": "http://creativecommons.org/licenses/by/4.0/",
"primary_cat": "cs.CL",
"submitted_at": "2026-05-04T18:17:05Z",
"title_canon_sha256": "d3761b2174bae54d4638480ef8a504bd600dacbb661389e838b37b4ae07ecbaa"
},
"schema_version": "1.0",
"source": {
"id": "2605.03052",
"kind": "arxiv",
"version": 2
}
}