Datasets:

Acture
/

MotifBench

@@ -58,3 +58,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 # Video files - compressed
 *.mp4 filter=lfs diff=lfs merge=lfs -text
 *.webm filter=lfs diff=lfs merge=lfs -text

 # Video files - compressed
 *.mp4 filter=lfs diff=lfs merge=lfs -text
 *.webm filter=lfs diff=lfs merge=lfs -text
+assets/EMNLP[[:space:]]2026_Find-1836.pdf filter=lfs diff=lfs merge=lfs -text

README.md CHANGED Viewed

@@ -1,171 +1,98 @@
 ---
 license: cc-by-nc-4.0
 language:
-  - en
 tags:
-  - long-context
-  - llm-evaluation
-  - graph-neural-networks
-  - reasoning
-  - literary-analysis
-pretty_name: MotifBench — Literary Motif Recovery
 size_categories:
-  - n<1K
 configs:
-  - config_name: macbeth
-    data_files:
-      - split: train
-        path: macbeth/macbeth.jsonl
-    default: true
 ---
 # MotifBench
-**Can a model connect an ocean in Act II with perfume in Act V?**
-A dataset for recovering recurring literary motifs from *Macbeth*, using associations extracted from published criticism. Accompanies **Critic’s Eye**, accepted to **Findings of EMNLP 2026**.
-[Quick start](#quick-start) · [Data and fields](#what-is-in-this-repository) · [Paper and citation](#paper-and-citation) · [Code](https://github.com/Acture/critics-eye)
-## Different images. A recurring idea.
-<img src="assets/ocean-perfume.png" alt="An ocean streaked with red and an ornate perfume bottle: two different images connected by the attempt to cleanse a bloody hand." width="760" />
-| Macbeth · Act II | Lady Macbeth · Act V |
-| --- | --- |
-| “Will all great Neptune’s **ocean wash this blood** / Clean from my hand?” | “the **perfumes of Arabia will not sweeten** this little hand.” |
-| Water cannot wash it away. | Perfume cannot cover it up. |
-Both passages return to **cleansing that fails**: a way of expressing guilt that cannot be erased. The imagery, speaker and scene change; the underlying connection remains. A **motif** is an element that recurs with variation across a work.
-Finding the passages is only part of the problem. A model must also connect what they mean. This is the long-range reasoning challenge that motivates MotifBench: a large context window gives access to distant text, but does not by itself establish those connections.
-*This is an illustrative reading, not a scored model prediction or a labeled pair in this export. Read the passages in Folger’s [Act 2, scene 2](https://www.folger.edu/explore/shakespeares-works/macbeth/read/2/2/) and [Act 5, scene 1](https://www.folger.edu/explore/shakespeares-works/macbeth/read/5/1/).*
-## What does a model predict?
-The paper’s **motif recovery** task asks a model to rank a fixed inventory of motifs for a given line. The target is an association recorded in literary criticism.
-| Input | Output | Reference |
-| --- | --- | --- |
-| A query line and a candidate motif inventory, with the context allowed by the evaluation setting | A ranked list of motifs | Recorded line–motif associations |
-The benchmark uses criticism because the target is interpretive: critics have already connected passages and named recurring elements in published work. The extraction pipeline locates these assertions and resolves their references to the play. The bibliography contains **26 sources**; the dated entries range from **1765 to 2020**.
-These are **recorded positive associations, not exhaustive labels**. A missing association does not establish that an interpretation is wrong. Extraction, motif grouping and citation resolution can also introduce errors.
-## Quick start
-Install `huggingface_hub` in your Python environment. While this repository is private, use `hf auth login` with an account that has access.
 ```python
-import json
 from huggingface_hub import hf_hub_download
-path: str = hf_hub_download(
-    repo_id="Acture/MotifBench",
-    filename="macbeth/macbeth.jsonl",
-    repo_type="dataset",
-)
-with open(path, encoding="utf-8") as stream:
-    rows: list[dict[str, str | None]] = [json.loads(line) for line in stream]
-line_rows: list[dict[str, str | None]] = [
-    row for row in rows if row["node_type"] == "line"
-]
-print(f"{len(rows)} associations; {len(line_rows)} line associations")
-print(line_rows[0])
 ```
-The current file prints `645 associations; 585 line associations`. Multiple rows can refer to the same line, so **rows are not independent query lines**. Pin `revision` to a Hub commit when recording experiments.
-## What is in this repository?
-This Hub snapshot is a **flattened association export**. It contains the following files:
-| File | Contents |
-| --- | --- |
-| [macbeth/macbeth.jsonl](macbeth/macbeth.jsonl) | Narrative-node–motif associations, with text for line nodes |
-| [macbeth/criticism_sources.csv](macbeth/criticism_sources.csv) | Source bibliography, publication information, access links and public-domain status as recorded in the release |
-| [LICENSE](LICENSE) | CC BY-NC 4.0 license |
-Counts below were computed directly from the data at revision `fa060c5ad43ea3fdccb7cb6ad02a7bc85fa20c70`. The September 2026 card update leaves those data files unchanged.
-| `node_type` | Association rows | Distinct source nodes |
-| --- | ---: | ---: |
-| `line` | 585 | 314 |
-| `scene` | 44 | 14 |
-| `stage_direction` | 7 | 5 |
-| `theme` | 9 | 9 |
-| **Total** | **645** | **342** |
-Across all rows, **277 distinct `motif_id` values** appear. These are the motifs represented as association targets in this file, not a separately distributed full candidate inventory.
-### Fields
-All fields are strings or `null`.
-| Field | Meaning |
-| --- | --- |
-| `node_type` | Source node type: `line`, `scene`, `stage_direction` or `theme` |
-| `node_id` | Source node identifier. Line IDs use Folger Through-Line Numbers, such as `ftln-1616`. |
-| `act_id`, `scene_id`, `section_id` | Available structural identifiers. A section denotes a speaking turn. Non-line rows have null values in these columns. |
-| `speaker` | Speaker identifier, where available |
-| `text` | Folger line text; null for non-line rows |
-| `motif_id` | Target motif identifier; an export ID, not a ranked position |
-| `motif_descriptor` | Text description associated with that motif in this export |
-### An actual row
-This record associates a line describing Banquo’s ghost with a motif descriptor. Macbeth is the **speaker**; Banquo is the person described.
-```json
-{
-  "node_type": "line",
-  "node_id": "ftln-1616",
-  "act_id": "act_4",
-  "scene_id": "act_4_scene_1",
-  "section_id": "sp-1605",
-  "speaker": "Macbeth_Mac",
-  "text": "For the blood-boltered Banquo smiles upon me",
-  "motif_id": "motif_74",
-  "motif_descriptor": "the ghost’s appearance described as ‘blood���baltered’ (4.1.138)"
-}
-```
-The descriptor is shown exactly as stored. It is not a model prediction.
-## Using the data for evaluation
-**This export is not a complete reproduction package for the EMNLP evaluation.** The paper’s frozen evaluation uses a 281-motif inventory and 351 membership annotations over 283 query lines. The Hub file above has a different population and does not preserve the relation types needed to isolate that membership set.
-The file also does not include the full narrative graph, evaluation fold assignments, per-association source citations, or a separate complete motif inventory. `criticism_sources.csv` is a bibliography, not a row-level provenance table. The Hub’s `train` split name is a loading convention for the single file; it does not define the paper’s training partition.
-For exploratory work with this export:
-- Filter to `node_type == "line"` for line-level analyses, and group associations by `node_id` before defining queries.
-- Keep all associations for the same query line together when constructing a split.
-- Record the data revision, candidate inventory, split, available context and scoring rule. Results under a new protocol should be identified as such.
-- Interpret unrecorded associations as unobserved, rather than automatically false.
-The released text covers the associated lines, not the complete play. Consult the [Folger edition](https://www.folger.edu/explore/shakespeares-works/macbeth/read/) when full narrative context is needed.
-## Text, sources and license
-The dataset is licensed **CC BY-NC 4.0**; see [LICENSE](LICENSE).
-Line text comes from the Folger Shakespeare Library’s digital edition of *Macbeth*, edited by Barbara A. Mowat and Paul Werstine, with encoding by Michael Poston and Rebecca Niles. The source edition is licensed **CC BY-NC 3.0**. Preserve this attribution when reusing the text.
-The repository supplies bibliographic metadata and available access links for the criticism sources. It does not host their full texts or PDFs. The `public_domain` column records the release’s source classification; one entry is `unclear`.
-The opening illustration is reused from the authors’ EMNLP poster. It is explanatory artwork, not experimental evidence; see [asset provenance](assets/README.md).
-## Paper and citation
-**Critic’s Eye: Review-Grounded Graph Reasoning for Long-Form Motif Discovery**
-Xinyu Sun · Zhangyue Yin · Xin Wang · Fudan University
-Accepted to Findings of EMNLP 2026 · Paper 1836-FIND
 ```bibtex
 @inproceedings{sun2026criticseye,
@@ -176,4 +103,6 @@ Accepted to Findings of EMNLP 2026 · Paper 1836-FIND
 }
 ```
-Contact: [sunxy23@m.fudan.edu.cn](mailto:sunxy23@m.fudan.edu.cn).

 ---
 license: cc-by-nc-4.0
 language:
+- en
 tags:
+- long-context
+- llm-evaluation
+- graph-neural-networks
+- reasoning
+pretty_name: MotifBench
 size_categories:
+- 1K<n<10K
 configs:
+- config_name: macbeth
+  data_files: macbeth/macbeth.jsonl
+  default: true
 ---
 # MotifBench
+[![Critic’s Eye — EMNLP 2026 poster](assets/EMNLP%202026_Find-1836.png)](https://huggingface.co/datasets/Acture/MotifBench/resolve/main/assets/EMNLP%202026_Find-1836.pdf)
+[Open the full-resolution poster (PDF)](https://huggingface.co/datasets/Acture/MotifBench/resolve/main/assets/EMNLP%202026_Find-1836.pdf)
+**A benchmark for whether models can assemble meaning that no single passage
+states — not just retrieve it.**
+[![Code](https://img.shields.io/badge/code-critics--eye-blue)](https://github.com/Acture/critics-eye)
+[![License](https://img.shields.io/badge/license-CC%20BY--NC%204.0-lightgrey)](LICENSE)
+## Why this benchmark
+Long-context benchmarks test **locating**: the answer sits somewhere in the
+input, and the task is to find it. Needle-in-a-haystack, multi-hop QA,
+book-length summarization — in all of them, retrieval eventually wins.
+MotifBench tests something retrieval cannot do. A **motif** is a recurring
+salient thematic element a work returns to — blood in *Macbeth*, surfacing
+in lines that share no vocabulary:
+> "the multitudinous seas incarnadine" (2.2) · "Out, damned spot!" (5.1) ·
+> "all the perfumes of Arabia" (5.1)
+None of these lines contains the word *blood*. No line states the motif —
+it exists only across them. Recovering it means noticing that scattered,
+dissimilar passages belong together: **assembly, not search**.
+The labels are not ours to invent. An interpretive target has no annotator
+we could trust more than the alternative, so MotifBench draws its ground
+truth from published literary criticism (1765–2020) — critics who named
+these motifs, in print, long before any benchmark existed. A language model
+is used only to *locate* where a critic asserts a link and resolve the
+citation; nothing enters the benchmark on the model's word.
+Paper: *Critic's Eye: Review-Grounded Graph Reasoning for Long-Form Motif
+Discovery* (EMNLP 2026). Code: <https://github.com/Acture/critics-eye>.
+## Dataset structure (`macbeth/`)
+| file | contents |
+|---|---|
+| `macbeth.jsonl` | One row per (narrative node, motif) association: `node_type`, `node_id`, `act_id`, `scene_id`, `section_id`, `speaker`, `text`, `motif_id`, `motif_descriptor`. `node_type` is `line`, `scene`, or `stage_direction`; `node_id` for a line is the Folger Through-Line-Number (FTLN), e.g. `ftln-0001`. `text` holds the Folger line text for `line` rows and is `null` for `scene`/`stage_direction` rows. |
+| `criticism_sources.csv` | The 26 criticism works this benchmark draws on: author, year, title, venue, public-domain status, access URL/DOI. |
+### Example row
+```json
+{"node_type": "line", "node_id": "ftln-0004", "act_id": "act_1", "scene_id": "act_1_scene_1", "section_id": "sp-0003", "speaker": "WITCHES.2_Mac", "text": "When the battle’s lost and won.", "motif_id": "motif_23", "motif_descriptor": "lady macbeth's line \"wouldst not play false, / and yet wouldst wrongly win\""}
+```
+## Usage
 ```python
+import pandas as pd
 from huggingface_hub import hf_hub_download
+path = hf_hub_download(repo_id="Acture/MotifBench", filename="macbeth/macbeth.jsonl", repo_type="dataset")
+df = pd.read_json(path, lines=True)
+print(df[["node_id", "text", "motif_id", "motif_descriptor"]].head())
 ```
+## Text and copyright
+`text` on `line` rows is from the Folger Shakespeare Library's digital
+edition of *Macbeth* (Barbara A. Mowat and Paul Werstine, eds.; encoded by
+Michael Poston and Rebecca Niles; <https://www.folgerdigitaltexts.org>),
+licensed CC BY-NC 3.0.
+Of the 26 criticism works, public-domain ones (pre-1922) are pointed to in
+full (mostly via Project Gutenberg); copyrighted ones are referenced by
+metadata only — see `criticism_sources.csv`.
+## Citation
+Provisional BibTeX; the final ACL Anthology entry is pending.
 ```bibtex
 @inproceedings{sun2026criticseye,
 }
 ```
+## License
+CC BY-NC 4.0 — see `LICENSE`.

assets/EMNLP 2026_Find-1836.pdf ADDED Viewed

	@@ -0,0 +1,3 @@

+version https://git-lfs.github.com/spec/v1
+oid sha256:a6e011f0c2fe265177497bd5e2635bed43f6bc5c55f594eecbed5fde13cd67c6
+size 16296493

assets/{ocean-perfume.png → EMNLP 2026_Find-1836.png} RENAMED Viewed

File without changes

assets/README.md DELETED Viewed

@@ -1,7 +0,0 @@
-# Illustration provenance
-`ocean-perfume.png` is an unchanged copy of the authors’ EMNLP poster asset `acl/poster/assets/motif_objects_transparent_v17.png` in the Critic’s Eye paper repository.
-The artwork was generated with OpenAI image generation and subsequently edited with the same tool to remove its background. It depicts an ocean and a perfume bottle as an illustration of the Macbeth reading described in the dataset card. It is not a historical image, a source-document scan, a benchmark annotation, or a measured model response.
-The original resolution is 1774 × 887 pixels, with an alpha channel. No new image generation or image editing was performed for this card update.