---
title: "Evidence Grounding vs. Memorization: Why Neural Semantics"
description: "The authors compare these under one pipeline on FACTKG, a DBpedia benchmark of 108,675 claims labelled with reasoning… 24 references."
authors: ["Ankit Kumar Upadhyay", "John S. Erickson", "Deborah L. McGuinness"]
year: 2026
venue: "Proceedings of the Ninth Fact Extraction and VERification Workshop (FEVER)"
source: "https://aclanthology.org/2026.fever-1.3/"
doi: "10.18653/v1/2026.fever-1.3"
licence: "CC BY 4.0 (ACL Anthology)"
licence_url: "https://creativecommons.org/licenses/by/4.0/"
summary_licence: "Elamur Cosmic License"
canonical: "https://research.elamur.ai/free/en/papers/upadhyay-2026-evidence-grounding-vs-memorization-why/"
---
[Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/papers/upadhyay-2026-evidence-grounding-vs-memorization-why/)

[home](https://md.research.elamur.ai/free/en/) / [papers](https://md.research.elamur.ai/free/en/papers/) / [2026](https://md.research.elamur.ai/free/en/papers/2026/page/32/) / Evidence Grounding vs. Memorization: Why Neural Semantics Matter for Knowledge Graph Fact Verification

# Evidence Grounding vs. Memorization: Why Neural Semantics Matter for Knowledge Graph Fact Verification

**Authors:** Ankit Kumar Upadhyay, John S. Erickson and Deborah L. McGuinness · **Year:** 2026 · **Venue:** [Proceedings of the Ninth Fact Extraction and VERification Workshop (FEVER)](https://md.research.elamur.ai/free/en/venues/fever/) ([FEVER 2026](https://md.research.elamur.ai/free/en/venues/fever/2026/)) · **Publisher:** [Association for Computational Linguistics](https://md.research.elamur.ai/free/en/publishers/association-for-computational-linguistics/) · **Licence:** [CC BY 4.0 (ACL Anthology)](https://creativecommons.org/licenses/by/4.0/)

**Links:** [Publisher's page](https://aclanthology.org/2026.fever-1.3/) · [PDF](https://aclanthology.org/2026.fever-1.3.pdf) · [DOI 10.18653/v1/2026.fever-1.3](https://doi.org/10.18653/v1/2026.fever-1.3) · [OpenAlex](https://openalex.org/W7140148821)

**Citations:** 0 citations (OpenAlex, 2026-10-14)

## Summary
Verifying a natural-language claim against a knowledge graph requires deciding whether structured triples support or contradict it. It is unclear how much success comes from symbolic structure, neural text semantics, or explicit evidence. The authors compare these under one pipeline on FACTKG, a DBpedia benchmark of 108,675 claims labelled with reasoning types such as negation, existence and multi-hop. They test an XGBoost model over 29 hand-engineered features, a reproduced BERT-base model reading linearized one-hop subgraphs, and QA-GNN-style graph networks. They also use GPT-4.1-mini to choose the ten most relevant triples per claim before fine-tuning BERT. Finally, two GPT models are evaluated with and without KG evidence on a stratified 300-claim sample. BERT clearly leads at 92.68%, far ahead of symbolic features and GNNs. LLM-based triple selection makes training data more learnable, and supplying subgraphs with chain-of-thought lifts LLM accuracy by roughly 9–13 points. The study suggests that token-level semantics and citable evidence matter more than memorized knowledge or message passing.

## Key points
- On FACTKG (86,367 / 13,267 / 9,041 train/validation/test split), the symbolic XGBoost baseline reached 66.54%. It performed worst on negation (41.10%), and GNN variants levelled off near 70%.
- The reproduced BERT-base over linearized one-hop subgraphs scored 92.68%, close to Opsahl's earlier 93.49%. Subgraphs average over 50 triples and often exceed the 512-token limit.
- On 9,645 training examples, BERT reached 78.85% with GPT-4.1-mini top-k=10 triple filtering. It reached 77.54% with lexical heuristic selection and only 52.70% with plain truncation.
- With KG grounding, GPT-4o-mini rose from 71.67% to 84.33% and GPT-4.1-mini from 74.67% to 84.00%. Responses cited on average 6.28 of 36.7 available triples.
- Multi-hop claims did not benefit from grounding, staying at 73.03% for GPT-4.1-mini in both modes. Grounded accuracy was also much higher on REFUTED (93.79%) than on SUPPORTED (69.92%) claims.

*Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.*

## Institutions

- [Rensselaer Polytechnic Institute](https://md.research.elamur.ai/free/en/universities/rensselaer-polytechnic-institute/) — University, Troy, United States · 23 works · 5,092 citations

📚 Shelves (6): [🖥️ Computer Science](https://md.research.elamur.ai/free/en/shelves/computer-science/) · 18,614 works · [🤖 Artificial Intelligence](https://md.research.elamur.ai/free/en/shelves/computer-science/artificial-intelligence/) · 17,965 works · [🎭 Arts and Humanities](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/) · 17,450 works · [🗣️ Language and Linguistics](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/language-and-linguistics/) · 17,296 works · [🤝 Social Sciences](https://md.research.elamur.ai/free/en/shelves/social-sciences/) · 1,954 works · [🤝 Sociology and Political Science](https://md.research.elamur.ai/free/en/shelves/social-sciences/sociology-and-political-science/) · 626 works

🧭 Topics (1): [🕸️ Knowledge graphs & ontologies](https://md.research.elamur.ai/free/en/topics/knowledge-graphs/) · 211 works

## Cite

Cite as: Ankit Kumar Upadhyay, John S. Erickson and Deborah L. McGuinness. 2026. Evidence Grounding vs. Memorization: Why Neural Semantics Matter for Knowledge Graph Fact Verification. Proceedings of the Ninth Fact Extraction and VERification Workshop (FEVER).

The paper itself, not this page. BibTeX from the library's record of the work.

```bibtex
@inproceedings{upadhyay2026evidence,
  title = {{Evidence Grounding vs. Memorization: Why Neural Semantics Matter for Knowledge Graph Fact Verification}},
  author = {Ankit Kumar Upadhyay, John S. Erickson and Deborah L. McGuinness},
  booktitle = {Proceedings of the Ninth Fact Extraction and VERification Workshop (FEVER)},
  year = {2026},
  doi = {10.18653/v1/2026.fever-1.3},
  url = {https://doi.org/10.18653/v1/2026.fever-1.3}
}
```

- [← previous EVGeoQA: Benchmarking LLMs on Dynamic, Multi-Objective Geo-Spatial Exploration](https://md.research.elamur.ai/free/en/papers/wu-2026-evgeoqa-benchmarking-llms-dynamic-multi/)
- [next → Evidence-Augmented Generation Reasoning for Extremely Low-Resource Language Decipherment](https://md.research.elamur.ai/free/en/papers/zhu-2026-evidence-augmented-generation-reasoning-extremely/)
- [↑ back to papers of 2026](https://md.research.elamur.ai/free/en/papers/2026/page/32/)

## References

24 references, 11 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

1. Jiangui Chen, Ruqing Zhang, Jiafeng Guo, Yixing Fan, and Xueqi Cheng. 2022. Gere: Generative evidence retrieval for fact verification. In Proceedings of the 45th International ACM SIGIR Conference on Research and Development in Information Retrieval, pages 2184–2189. [find](https://scholar.google.com/scholar?q=Jiangui+Chen%2C+Ruqing+Zhang%2C+Jiafeng+Guo%2C+Yixing+Fan%2C+and+Xueqi+Cheng%2E+2022%2E+Gere%3A+Generative+evidence+retrieval+for+fact+verification%2E+In+Proceedings+of+the+45th+International+ACM+SIGIR+Conference+on+)
2. Wenhu Chen, Hongmin Wang, Jianshu Chen, Yunkai Zhang, Hong Wang, Shiyang Li, Xiyou Zhou, and William Yang Wang. 2019. [Tabfact: A large-scale dataset for table-based fact verification](https://md.research.elamur.ai/free/en/papers/chen-2019-tabfact-large-scale-dataset-table/). arXiv preprint arXiv:1909.02164.
3. Lorenzo De Felice, Elena Cabrio, and Serena Villata. 2025. Evoca: Explainable verification of claims by graph alignment. Information, 16(1):45. [find](https://scholar.google.com/scholar?q=Lorenzo+De+Felice%2C+Elena+Cabrio%2C+and+Serena+Villata%2E+2025%2E+Evoca%3A+Explainable+verification+of+claims+by+graph+alignment%2E+Information%2C+16%281%29%3A45%2E)
4. Christopher Hidey, Tuhin Chakrabarty, Tariq Alhindi, Siddharth Varia, Kriste Krstovski, Mona Diab, and Smaranda Muresan. 2020. Deseption: Dual sequence prediction and adversarial examples for improved fact-checking. arXiv preprint arXiv:2004.12864. [find](https://scholar.google.com/scholar?q=Christopher+Hidey%2C+Tuhin+Chakrabarty%2C+Tariq+Alhindi%2C+Siddharth+Varia%2C+Kriste+Krstovski%2C+Mona+Diab%2C+and+Smaranda+Muresan%2E+2020%2E+Deseption%3A+Dual+sequence+prediction+and+adversarial+examples+for+improved)
5. Jiho Kim, Sungjin Park, Yeonsu Kwon, Yohan Jo, James Thorne, and Edward Choi. 2023. Factkg: Fact verification via reasoning on knowledge graphs. In Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics, pages 16190– 16206. [find](https://scholar.google.com/scholar?q=Jiho+Kim%2C+Sungjin+Park%2C+Yeonsu+Kwon%2C+Yohan+Jo%2C+James+Thorne%2C+and+Edward+Choi%2E+2023%2E+Factkg%3A+Fact+verification+via+reasoning+on+knowledge+graphs%2E+In+Proceedings+of+the+61st+Annual+Meeting+of+the+Associ)
6. Jens Lehmann, Robert Isele, Max Jakob, Anja Jentzsch, Dimitris Kontokostas, Pablo N Mendes, Sebastian Hellmann, Mohamed Morsey, Patrick van Kleef, Sören Auer, and Christian Bizer. 2015. [Dbpedia– a large-scale, multilingual knowledge base extracted from wikipedia](https://md.research.elamur.ai/free/en/papers/lehmann-2015-dbpedia-large-scale-multilingual-knowledge/). Semantic Web, 6(2):167–195.
7. Jason Wei, Xuezhi Wang, Dale Schuurmans, Maarten Bosma, Brian Ichter, Fei Xia, Ed H Chi, Quoc V Le, and Denny Zhou. 2022. [Chain-of-thought prompting elicits reasoning in large language models](https://md.research.elamur.ai/free/en/papers/wei-2022-chain-thought-prompting-elicits-reasoning/). In Advances in Neural Information Processing Systems, volume 35, pages 24824–24837.
8. Jiayuan Mao, Chuang Gan, Pushmeet Kohli, Joshua B Tenenbaum, and Jiajun Wu. 2019. [The neuro-symbolic concept learner: Interpreting scenes, words, and sentences from natural supervision](https://md.research.elamur.ai/free/en/papers/mao-2019-neuro-symbolic-concept-learner-interpreting/). In International Conference on Learning Representations.
9. Michihiro Yasunaga, Antoine Bosselut, Hongyu Ren, Xikun Zhang, Christopher D Manning, Percy Liang, and Jure Leskovec. 2022. Deep bidirectional language-knowledge graph pretraining. In Advances in Neural Information Processing Systems. [find](https://scholar.google.com/scholar?q=Michihiro+Yasunaga%2C+Antoine+Bosselut%2C+Hongyu+Ren%2C+Xikun+Zhang%2C+Christopher+D+Manning%2C+Percy+Liang%2C+and+Jure+Leskovec%2E+2022%2E+Deep+bidirectional+language%2Dknowledge+graph+pretraining%2E+In+Advances+in+Neur)
10. Shinya Momii, Naoya Inoue, and Kentaro Inui. 2023. Rule-based fact verification utilizing knowledge graphs. In Proceedings of the 6th Joint International Conference on Data Science & Management of Data. [find](https://scholar.google.com/scholar?q=Shinya+Momii%2C+Naoya+Inoue%2C+and+Kentaro+Inui%2E+2023%2E+Rule%2Dbased+fact+verification+utilizing+knowledge+graphs%2E+In+Proceedings+of+the+6th+Joint+International+Conference+on+Data+Science+%26+Management+of+Dat)
11. Michihiro Yasunaga, Hongyu Ren, Antoine Bosselut, Percy Liang, and Jure Leskovec. 2021. Qa-gnn: Reasoning with language models and knowledge graphs for question answering. In Proceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics, pages 535–… [find](https://scholar.google.com/scholar?q=Michihiro+Yasunaga%2C+Hongyu+Ren%2C+Antoine+Bosselut%2C+Percy+Liang%2C+and+Jure+Leskovec%2E+2021%2E+Qa%2Dgnn%3A+Reasoning+with+language+models+and+knowledge+graphs+for+question+answering%2E+In+Proceedings+of+the+2021+C)
12. Mark Newman. 2018. Networks, 2nd edition. Oxford University Press, Oxford. [find](https://scholar.google.com/scholar?q=Mark+Newman%2E+2018%2E+Networks%2C+2nd+edition%2E+Oxford+University+Press%2C+Oxford%2E)
13. Yixin Nie, Haonan Chen, and Mohit Bansal. 2019. Combining fact extraction and verification with neural semantic matching networks. In Proceedings of the AAAI conference on artificial intelligence, volume 33, pages 6859–6866. [find](https://scholar.google.com/scholar?q=Yixin+Nie%2C+Haonan+Chen%2C+and+Mohit+Bansal%2E+2019%2E+Combining+fact+extraction+and+verification+with+neural+semantic+matching+networks%2E+In+Proceedings+of+the+AAAI+conference+on+artificial+intelligence%2C+vol)
14. Kexin Yi, Jiajun Wu, Chuang Gan, Antonio Torralba, Pushmeet Kohli, and Joshua B Tenenbaum. 2018. [Neural-symbolic vqa: Disentangling reasoning from vision and language understanding](https://md.research.elamur.ai/free/en/papers/yi-2018-neural-symbolic-vqa-disentangling-reasoning/). In Advances in Neural Information Processing Systems, pages 1039– 1050.
15. Tobias A Opsahl. 2024. Fact or fiction? Improving fact verification with knowledge graphs through simplified subgraph retrievals. In Proceedings of the 6th Workshop on Fact Extraction and VERification (FEVER), pages 289–298. [find](https://scholar.google.com/scholar?q=Tobias+A+Opsahl%2E+2024%2E+Fact+or+fiction%3F+Improving+fact+verification+with+knowledge+graphs+through+simplified+subgraph+retrievals%2E+In+Proceedings+of+the+6th+Workshop+on+Fact+Extraction+and+VERification)
16. Zhaowei Yuan and Andreas Vlachos. 2024. Zero-shot fact verification with semantic triples and knowledge graphs. In Findings of the Association for Computational Linguistics: ACL 2024, pages 12844–12865. [find](https://scholar.google.com/scholar?q=Zhaowei+Yuan+and+Andreas+Vlachos%2E+2024%2E+Zero%2Dshot+fact+verification+with+semantic+triples+and+knowledge+graphs%2E+In+Findings+of+the+Association+for+Computational+Linguistics%3A+ACL+2024%2C+pages+12844%E2%80%931286)
17. Liangming Pan, Xiaobao Wu, Xinyuan Lu, Anh Tuan Luu, William Yang Wang, Min-Yen Kan, and Preslav Nakov. 2023. [Fact-checking complex claims with program-guided reasoning](https://md.research.elamur.ai/free/en/papers/pan-2023-fact-checking-complex-claims-program/). In Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics, pages 6981–7004.
18. Xikun Zhang, Antoine Bosselut, Michihiro Yasunaga, Hongyu Ren, Percy Liang, Christopher D Manning, and Jure Leskovec. 2021. Greaselm: Graph reasoning enhanced language models. In International Conference on Learning Representations. [find](https://scholar.google.com/scholar?q=Xikun+Zhang%2C+Antoine+Bosselut%2C+Michihiro+Yasunaga%2C+Hongyu+Ren%2C+Percy+Liang%2C+Christopher+D+Manning%2C+and+Jure+Leskovec%2E+2021%2E+Greaselm%3A+Graph+reasoning+enhanced+language+models%2E+In+International+Confere)
19. Hannah Rashkin, Vitaly Nikolaev, Matthew Lamm, Michael Collins, Dipanjan Das, Slav Petrov, Gaurav Singh Tomar, Iulia Turc, and David Reitter. 2023. [Measuring attribution in natural language generation models](https://md.research.elamur.ai/free/en/papers/rashkin-2023-measuring-attribution-natural-language-generation/). Computational Linguistics, 49(4):777–840.
20. Jie Zhou, Xu Han, Cheng Yang, Zhiyuan Liu, Lifeng Wang, Changcheng Li, and Maosong Sun. 2019. [Gear: Graph-based evidence aggregating and reasoning for fact verification](https://md.research.elamur.ai/free/en/papers/zhou-2019-gear-graph-based-evidence-aggregating/). arXiv preprint arXiv:1908.01843.
21. Mikhail Salnikov, Le Fang, Aniko Hannak, Hakan Ferhatosmanoglu, and Artem Baklanov. 2023. Large language models meet knowledge graphs to answer factoid questions. In Findings of the Association for Computational Linguistics: EMNLP 2023. [find](https://scholar.google.com/scholar?q=Mikhail+Salnikov%2C+Le+Fang%2C+Aniko+Hannak%2C+Hakan+Ferhatosmanoglu%2C+and+Artem+Baklanov%2E+2023%2E+Large+language+models+meet+knowledge+graphs+to+answer+factoid+questions%2E+In+Findings+of+the+Association+for+Co)
22. Tal Schuster, Adam Fisch, and Regina Barzilay. 2021. [Get your vitamin c! robust fact verification with contrastive evidence](https://md.research.elamur.ai/free/en/papers/schuster-2021-get-your-vitamin-c-robust/). arXiv preprint arXiv:2103.08541.
23. Tal Schuster, Darsh J Shah, Yun Jie Serene Yeo, Daniel Filizzola, Enrico Santus, and Regina Barzilay. 2019. [Towards debiasing fact verification models](https://md.research.elamur.ai/free/en/papers/schuster-2019-debiasing-fact-verification-models/). arXiv preprint arXiv:1908.05267.
24. James Thorne, Andreas Vlachos, Christos Christodoulopoulos, and Arpit Mittal. 2018. [Fever: a large-scale dataset for fact extraction and verification](https://md.research.elamur.ai/free/en/papers/thorne-2018-fever-large-scale-dataset-fact/). In Proceedings of the 2018 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pa…

*Updated 2 November 2026*