Find It Fix It
FIFI 2026

Shared Task on Scientific Title Retrieval & Rewriting — can your system trace an LLM-rewritten title back to its source?


⚙ Technical

"A Transformer-Based Framework for Cross-Domain Semantic Title Alignment"

📖 Accessible

"How AI Learns to Match Research Topics Across Different Fields"

🔥 Catchy

"Lost in Translation? AI Finally Bridges the Gap Between Research Silos"

Registration Form Submission Form Explore the Task → 📅 Key Dates

📢 Supporting Dataset for Task 1 Released

📢 The sample submission TSV file is available under the Evaluation tab. Please download and refer to it before submitting your final file.
📢 The submission format for Task 2 is the same as the validation file format. Please follow the validation file format while preparing your submission.
3
Style categories
2
Subtasks
10
Variants per style

What is FIFI?

LLMs can rewrite scientific titles into three styles — Technical, Accessible, and Catchy. FIFI challenges participants to reverse that transformation: given only a reformulated title, retrieve or reconstruct the original research title.

🔍

Subtask 1: Find It

Rank up to 10 candidate titles by likelihood of being the true source. Evaluated with MRR@10.

✍️

Subtask 2: Fix It

Reconstruct the original title from its stylistic variant. Evaluated with F1 Score & BERTScore.