Pith Number

pith:BX2RKHOZ

pith:2026:BX2RKHOZEXXP62EF4BIIQW454G

not attested not anchored not stored refs pending

Why Are Some Emotions Harder for LLMs? Uncovering the Causal Mechanisms of Emotion Inference via Sparse Autoencoders

Arinjay Singh, Bangzhao Shu, Mai ElSherief

LLMs process emotions in a distinct final phase using features that can be adjusted to improve recognition while keeping language abilities intact.

arxiv:2604.25866 v2 · 2026-04-28 · cs.CL

Open paper page JSON Open Graph Bundle Merged state Verified badge What is a Pith Number?

Add to your LaTeX paper

\usepackage{pith}
\pithnumber{BX2RKHOZEXXP62EF4BIIQW454G}

Prints a linked badge after your title and injects PDF metadata. Compiles on arXiv. Learn more · Embed verified badge

Record completeness

1 Bitcoin timestamp

2 Internet Archive

3 Author claim open · sign in to claim

4 Citations open

5 Replications open

✓ Portable graph bundle live · download bundle · merged state

The bundle contains the canonical record plus signed events. A mirror can host it anywhere and recompute the same current state with the deterministic merge algorithm.

Claims

C1strongest claim

we identify a consistent three-phase information flow, in which emotion-related features emerge only in the final phase... propose an interpretable and data-efficient causal feature steering method that significantly improves emotion recognition performance across multiple models while largely preserving language modeling ability, and demonstrate that these improvements generalize across multiple emotion recognition datasets.

C2weakest assumption

That the sparse features recovered by autoencoders correspond to genuine, causally relevant emotion computations inside the LLM rather than artifacts of the SAE training or post-hoc selection.

C3one line summary

LLMs represent emotions through late-emerging shared and specific sparse features whose causal intervention improves recognition performance across models and datasets.

Receipt and verification

First computed	2026-06-26T01:15:52.798240Z
Builder	pith-number-builder-2026-05-17-v1
Signature	Pith Ed25519 (`pith-v1-2026-05`) · public key
Schema	pith-number/v1.0

Canonical hash

0df5151dd925eeff6885e050885b9de1b016018866321b0f9e5ac7705f54f32e

Aliases

arxiv: 2604.25866 · arxiv_version: 2604.25866v2 · doi: 10.48550/arxiv.2604.25866 · pith_short_12: BX2RKHOZEXXP · pith_short_16: BX2RKHOZEXXP62EF · pith_short_8: BX2RKHOZ

Agent API

Resolver JSON Graph JSON Events JSON Schema Signing key

Verify this Pith Number yourself

curl -sH 'Accept: application/ld+json' https://pith.science/pith/BX2RKHOZEXXP62EF4BIIQW454G \
  | 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: 0df5151dd925eeff6885e050885b9de1b016018866321b0f9e5ac7705f54f32e

Canonical record JSON

{
  "metadata": {
    "abstract_canon_sha256": "791f6fd82702e3de8b82d80d457795b3da5deff11fb48f9c43ef6de940a81c52",
    "cross_cats_sorted": [],
    "license": "http://creativecommons.org/licenses/by/4.0/",
    "primary_cat": "cs.CL",
    "submitted_at": "2026-04-28T17:03:16Z",
    "title_canon_sha256": "71a518710e30b09df28c9e764c9f780a163c483e93097c6d1e7c3a134492eebe"
  },
  "schema_version": "1.0",
  "source": {
    "id": "2604.25866",
    "kind": "arxiv",
    "version": 2
  }
}