Elamur Research⌕
☰
home / papers / 2026 / LeedsMEng26: Qwen + Gemini for FinCausal 2026 Causality Detection in Financial Narrative Texts

LeedsMEng26: Qwen + Gemini for FinCausal 2026 Causality Detection in Financial Narrative Texts

Paper · 2026 · Zaid Shahrouri, Ayomide Ivienagbor, Idrees Asad et al.

Summary

The paper describes the University of Leeds entry to the FinCausal 2026 shared task. The task asks a system to answer abstractive questions about causes or effects in English and Spanish financial report paragraphs. Every answer must be an exact span copied from the context. The authors build a modular two-stage system. First, a model proposes a candidate span. Then an instruction-following LLM checks that span and adjusts its boundaries. If the second model returns text that does not appear in the context, the system keeps the original candidate.

They compare several setups:

  • a fine-tuned RoBERTa extractor;
  • Qwen-2.5-1.5B-Instruct adapted with LoRA and explored with reinforcement learning;
  • refiners based on GPT-3.5, GPT-4.0 mini and Gemini-2.5-flash.

Qwen produced better candidates than RoBERTa, and adding a verifier improved results further. The strongest combination, Qwen paired with Gemini-2.5-flash, ranked 7th in the English track. The work shows that a small, cheaply trained generator plus an LLM checker can be competitive under tight compute limits.

Key points

  • Each FinCausal 2026 data file has 2000 rows of ID, context, question and answer. Scoring uses an LLM-as-a-judge adequacy scale from 1 to 5, which replaces the earlier SAS and EM metrics.
  • The best system, Qwen-2.5-1.5B-Instruct with Gemini-2.5-flash, scored 4.7000 adequacy in English and 4.6143 in Spanish.
  • The extractive baseline was question-answering-roberta-base-s-v2, fine-tuned on English training data. Its most common errors were cutting answers short and running past the correct boundary.
  • Qwen was trained with LoRA and RL post-training using a composite reward that combines EM, semantic similarity and a judge score. This was done on a single NVIDIA RTX 3050.
  • The authors list several limitations: training focused on English only, the dataset is small, time and GPU access were limited, and the newest proprietary APIs were unavailable.

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

🎓University of LeedsLeeds · United Kingdom40 works · 1,822 citations

 Fields, topics and concepts

📚 Shelves 5The field of science this work belongs to (OpenAlex fields).Computer Science · 16,667 worksArtificial Intelligence · 16,194 worksArts and Humanities · 16,039 worksLanguage and Linguistics · 15,891 worksComputer Vision and Pattern Recognition · 2,869 works

Cite

Cite as Zaid Shahrouri, Ayomide Ivienagbor, Idrees Asad, Rijul Shrestha, Yasemin Bal and Zahaab Nadeem. 2026. LeedsMEng26: Qwen + Gemini for FinCausal 2026 Causality Detection in Financial Narrative Texts. The 7th Financial Narrative Processing Workshop.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.63317/4aaj7cwvpjxy

@inproceedings{shahrouri2026leedsmeng26,
  title = {{LeedsMEng26: Qwen + Gemini for FinCausal 2026 Causality Detection in Financial Narrative Texts}},
  author = {Zaid Shahrouri, Ayomide Ivienagbor, Idrees Asad, Rijul Shrestha, Yasemin Bal and Zahaab Nadeem},
  booktitle = {The 7th Financial Narrative Processing Workshop},
  year = {2026},
  doi = {10.63317/4aaj7cwvpjxy},
  url = {https://doi.org/10.63317/4aaj7cwvpjxy}
}

References

10 references, 2 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. 1st Workshop on Large Language Models for Finance and Legal (LLMFinLegal), pages 214–221, Abu Dhabi, UAE. Association for Computational Linguistics. find
  2. Antonio Moreno-Sandoval, Jordi Porta, Yanco Torterolo, Alexia Stanescu, Melina Chatzi, and Sofía Roseti. 2026a. find
  3. The Financial Document Causality Detection Shared Task (FinCausal 2026). In Proceedings of the 7th Financial Narrative Processing Workshop (FNP 2026) at LREC 2026, Palma de Mallorca, Spain. ELRA.
  4. Antonio Moreno-Sandoval, Jordi Porta-Zamorano, Blanca Carbajo-Coronado, Doaa Samy, Dominique Mariko, and Mahmoud El-Haj. 2023. The financial document causality detection shared task (fincausal 2023). In 2023 IEEE International Conference on Big Data (BigData), Sorrento, Italy. IEEE. find
  5. Moreno-Sandoval, Antonio and Torterolo Orta, Yanco Amor and Stanescu, Maria Alexia and Chatzi, Melina. 2026b. find
  6. The Financial Document Causality Detection Shared Task (FinCausal 2026): Dataset. e-cienciaDatos.
  7. John Schulman and Thinking Machines Lab. 2025. Lora without regret. Thinking Machines Lab: Connectionism. find
  8. Avinash Trivedi, Gauri Toshniwal, Sivanesan Sangeetha, and S. R. Balasundaram. 2025. Sarang at FinCausal 2025: Contextual QA for financial causality detection combining extractive and generative models. In Proceedings of the Joint Workshop of the 9th Financial Technology and Natural Language Process… find
  9. Dominique Mariko, Kim Trottier, and Mahmoud El-Haj. 2022. The financial causality extraction shared task (fincausal 2022). In Proceedings of the 4th Financial Narrative Processing Workshop @ LREC 2022, pages 105–107, Marseille, France. European Language Resources Association. find
  10. Antonio Moreno-Sandoval, Blanca Carbajo Coronado, Jordi Porta Zamorano, Yanco Amor Torterolo Orta, and Doaa Samy. 2025. The financial document causality detection shared task (FinCausal 2025). In Proceedings of the Joint Workshop of the 9th Financial Technology and Natural Language Processing (FinNL… find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/shahrouri-2026-leedsmeng26-qwen-gemini-fincausal-2026/