Clarify model licensing and cite the AspectBench article

CITATION.cff

@@ -0,0 +1,28 @@
+cff-version: 1.2.0
+message: "If you use AspectBench, please cite the accompanying article."
+title: "AspectBench: Document-Level Aspect-Based Sentiment Analysis"
+type: software
+authors:
+  - family-names: Chatterjee
+    given-names: Nishan
+repository-code: "https://github.com/nishan-chatterjee/aspect-based-sentiment-analysis"
+url: "https://github.com/nishan-chatterjee/aspect-based-sentiment-analysis"
+preferred-citation:
+  type: article
+  authors:
+    - family-names: Chatterjee
+      given-names: Nishan
+    - family-names: Koloski
+      given-names: Boshko
+    - family-names: Doucet
+      given-names: Antoine
+    - family-names: Pollak
+      given-names: Senja
+    - family-names: Purver
+      given-names: Matthew
+  journal: "Frontiers in Artificial Intelligence"
+  volume: "9"
+  year: 2026
+  title: "Evaluating Fine-Tuned, Embedding-Based, and Zero-Shot Models for Aspect-Based Sentiment Analysis in South Slavic News"
+  doi: "10.3389/frai.2026.1844418"
+  url: "https://www.frontiersin.org/journals/artificial-intelligence/articles/10.3389/frai.2026.1844418/abstract"

LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2025 nishan
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

README.md

@@ -4,7 +4,7 @@ tags:
 - aspect-based-sentiment-analysis
 - south-slavic
 - text-classification
-license: other
+license: mit
 ---
 
 # AspectBench: reusable document-level ABSA inference
@@ -18,6 +18,30 @@ Carlo dropout.
 - [Reusable inference toolkit](https://huggingface.co/nishan-chatterjee/aspect-based-sentiment-analysis)
 - [GitHub repository](https://github.com/nishan-chatterjee/aspect-based-sentiment-analysis)
 
+## Licensing
+
+This repository contains the shared **MIT-licensed software toolkit**, not the
+fine-tuned weight files. Its license does not grant commercial rights to the
+separately downloaded AspectBench model contributions.
+
+The six non-Slavic-specific model families and the HBS BERTić checkpoint
+contributions are [CC BY-NC 4.0](https://creativecommons.org/licenses/by-nc/4.0/):
+noncommercial research and redistribution with attribution are allowed;
+commercial use requires separate permission. Consult each model card and
+`LICENSE` for exact scope and upstream attribution. Third-party base models,
+tokenizers and code retain their own licenses; the frozen BGE-M3 encoder remains
+upstream MIT while the trained AspectBench MLP heads are CC BY-NC 4.0.
+
+**Slovenian SloBERTa exception:** upstream EMBEDDIA/SloBERTa is CC BY-SA 4.0,
+which cannot simply be replaced with a noncommercial restriction. Those
+checkpoints are not relabeled CC BY-NC and need rights-holder/legal review to
+resolve the project's intended restriction. See the
+[Slavic-specific model card](https://huggingface.co/nishan-chatterjee/aspectbench-slavic-specific).
+
+Noncommercial is about the purpose of use, not academic/company affiliation.
+Dataset access terms are separate. Software remains MIT pending a separate
+licensing decision; earlier grants are not retroactively revoked.
+
 ## Input contract
 
 Every article must mark the target using literal `<aspect>...</aspect>` tags:
@@ -445,3 +469,21 @@ published only for independently trained variants.
 The public repositories are linked in the table near the top of this card and
 grouped in the AspectBench model collection. `scripts/download.py` restores the
 complete repository layout automatically.
+
+## Citation
+
+Please cite the accompanying article when using AspectBench:
+
+```bibtex
+@article{chatterjee2026aspectbench,
+  title   = {Evaluating Fine-Tuned, Embedding-Based, and Zero-Shot Models for
+             Aspect-Based Sentiment Analysis in South Slavic News},
+  author  = {Chatterjee, Nishan and Koloski, Boshko and Doucet, Antoine and
+             Pollak, Senja and Purver, Matthew},
+  journal = {Frontiers in Artificial Intelligence},
+  volume  = {9},
+  year    = {2026},
+  doi     = {10.3389/frai.2026.1844418},
+  url     = {https://www.frontiersin.org/journals/artificial-intelligence/articles/10.3389/frai.2026.1844418/abstract}
+}
+```

scripts/license_policy.py

@@ -0,0 +1,101 @@
+"""Release licensing and citation notices; never relicense upstream assets."""
+
+from __future__ import annotations
+
+import re
+
+from model_registry import MODEL_SPECS
+
+ARTICLE_URL = "https://www.frontiersin.org/journals/artificial-intelligence/articles/10.3389/frai.2026.1844418/abstract"
+NC_URL = "https://creativecommons.org/licenses/by-nc/4.0/"
+NC_LEGAL_URL = NC_URL + "legalcode.en"
+BIBTEX = """@article{chatterjee2026aspectbench,
+  title   = {Evaluating Fine-Tuned, Embedding-Based, and Zero-Shot Models for
+             Aspect-Based Sentiment Analysis in South Slavic News},
+  author  = {Chatterjee, Nishan and Koloski, Boshko and Doucet, Antoine and
+             Pollak, Senja and Purver, Matthew},
+  journal = {Frontiers in Artificial Intelligence},
+  volume  = {9},
+  year    = {2026},
+  doi     = {10.3389/frai.2026.1844418},
+  url     = {https://www.frontiersin.org/journals/artificial-intelligence/articles/10.3389/frai.2026.1844418/abstract}
+}"""
+
+# Public upstream card metadata inspected 2026-09-16. These are not new grants.
+UPSTREAM_LICENSES = {
+    "FacebookAI/xlm-roberta-base": "MIT",
+    "markussagen/xlm-roberta-longformer-base-4096": "Apache-2.0",
+    "microsoft/mdeberta-v3-base": "MIT",
+    "google/mt5-base": "Apache-2.0",
+    "classla/bcms-bertic": "Apache-2.0",
+    "EMBEDDIA/sloberta": "CC BY-SA 4.0",
+    "BAAI/bge-m3": "MIT",
+}
+
+
+def model_license_notice(model_name: str) -> str:
+    scope = (
+        "the HBS BERTić masked and unmasked checkpoint contributions only"
+        if model_name == "slavic-specific"
+        else "the AspectBench fine-tuned checkpoint contributions / trained heads"
+    )
+    text = f"""## License
+
+Copyright (c) 2026 the AspectBench model contributors.
+
+{scope[0].upper() + scope[1:]} are licensed under
+[Creative Commons Attribution-NonCommercial 4.0 International (CC BY-NC 4.0)]({NC_URL}).
+The [official legal code]({NC_LEGAL_URL}) is incorporated by reference.
+Noncommercial research, evaluation, adaptation and redistribution are allowed
+subject to attribution and the license terms. Commercial use is not granted
+under this license; contact the project for separate permission.
+
+Noncommercial describes the purpose of a use, not whether its user is a
+university or a company. Academic affiliation does not automatically make a
+commercial project noncommercial. The legal code controls, including its
+exceptions and limitations. The material is provided as-is without warranties.
+
+This notice does not relicense third-party base weights, tokenizers, code or
+configuration assets, and does not revoke any rights previously granted.
+AspectBench adapted the listed base models for aspect-based sentiment analysis;
+retain their original attribution and license notices when redistributing.
+"""
+    bases = MODEL_SPECS[model_name]["base_model"]
+    if isinstance(bases, str):
+        bases = {"HBS and Slovenian": bases}
+    for language, repo in bases.items():
+        text += f"\n- {language}: [{repo}](https://huggingface.co/{repo}) — upstream {UPSTREAM_LICENSES[repo]}."
+    if model_name == "slavic-specific":
+        text += """
+
+**Slovenian SloBERTa exception:** the upstream EMBEDDIA/SloBERTa model is
+CC BY-SA 4.0. No CC BY-NC grant or noncommercial restriction is applied here to
+the Slovenian checkpoints. Its ShareAlike terms cannot simply be replaced with
+CC BY-NC. Their licensing needs rights-holder/legal review before the project
+can promise a noncommercial-only release. Obtain an alternative upstream grant,
+use a compatible base model, or agree to retain ShareAlike (which permits
+commercial use). See the [upstream license](https://creativecommons.org/licenses/by-sa/4.0/legalcode.en).
+"""
+    if model_name == "bge-m3-mlp":
+        text += "\n\nThe frozen BGE-M3 encoder is downloaded separately and remains upstream MIT; the noncommercial grant covers the trained AspectBench MLP heads.\n"
+    return text.strip() + "\n"
+
+
+def citation_section() -> str:
+    return "## Citation\n\nPlease cite the accompanying article when using AspectBench:\n\n```bibtex\n" + BIBTEX + "\n```\n"
+
+
+def decorate_model_card(card: str, model_name: str) -> str:
+    """Add notices to generated cards as well as existing curated cards."""
+    metadata = (
+        "license: other\nlicense_name: aspectbench-language-specific\nlicense_link: LICENSE"
+        if model_name == "slavic-specific"
+        else "license: cc-by-nc-4.0"
+    )
+    card = re.sub(r"^license:.*$", metadata, card, count=1, flags=re.MULTILINE)
+    # Insert before first body section, immediately after the model introduction.
+    index = card.find("\n## ")
+    if index == -1:
+        index = len(card)
+    card = card[:index].rstrip() + "\n\n" + model_license_notice(model_name) + "\n" + card[index:].lstrip()
+    return card.rstrip() + "\n\n" + citation_section()

scripts/prepare_models.py

@@ -90,6 +90,12 @@ def export_state(source: Path, destination: Path, force: bool) -> dict[str, Any]
 
 
 def family_readme(model_name: str, entries: list[dict[str, Any]]) -> str:
+    from license_policy import decorate_model_card
+
+    return decorate_model_card(_family_readme(model_name, entries), model_name)
+
+
+def _family_readme(model_name: str, entries: list[dict[str, Any]]) -> str:
     spec = MODEL_SPECS[model_name]
     available = sum(item["available"] for item in entries)
     rows = []
@@ -355,6 +361,11 @@ def main() -> None:
         (family_dir / "README.md").write_text(
             family_readme(model_name, family_entries), encoding="utf-8"
         )
+        from license_policy import model_license_notice
+
+        (family_dir / "LICENSE").write_text(
+            model_license_notice(model_name), encoding="utf-8"
+        )
 
     manifest["available_slots"] = sum(
         entry["available"] for entry in manifest["entries"]