{"id":51504932,"url":"https://github.com/dcondrey/simpletext-clef2026","last_synced_at":"2026-07-07T23:01:26.293Z","repository":{"id":369432797,"uuid":"1249803429","full_name":"dcondrey/simpletext-clef2026","owner":"dcondrey","description":"CLEF 2026 SimpleText: multi-candidate LLM simplification (SARI 47.43) + DeBERTa/LightGBM overgeneration detection (2nd place, Task 2).","archived":false,"fork":false,"pushed_at":"2026-07-05T09:07:44.000Z","size":110,"stargazers_count":0,"open_issues_count":0,"forks_count":0,"subscribers_count":0,"default_branch":"main","last_synced_at":"2026-07-05T11:05:23.924Z","etag":null,"topics":["clef2026","deberta","hallucination-detection","lightgbm","llm","nlp","simpletext","text-simplification"],"latest_commit_sha":null,"homepage":null,"language":"Python","has_issues":true,"has_wiki":null,"has_pages":null,"mirror_url":null,"source_name":null,"license":"mit","status":null,"scm":"git","pull_requests_enabled":true,"icon_url":"https://github.com/dcondrey.png","metadata":{"files":{"readme":"README.md","changelog":null,"contributing":null,"funding":null,"license":"LICENSE","code_of_conduct":null,"threat_model":null,"audit":null,"citation":null,"codeowners":null,"security":null,"support":null,"governance":null,"roadmap":null,"authors":null,"dei":null,"publiccode":null,"codemeta":null,"zenodo":null,"notice":null,"maintainers":null,"copyright":null,"agents":null,"dco":null,"cla":null}},"created_at":"2026-05-26T03:44:16.000Z","updated_at":"2026-07-05T09:07:48.000Z","dependencies_parsed_at":null,"dependency_job_id":null,"html_url":"https://github.com/dcondrey/simpletext-clef2026","commit_stats":null,"previous_names":["dcondrey/simpletext-clef2026"],"tags_count":null,"template":false,"template_full_name":null,"purl":"pkg:github/dcondrey/simpletext-clef2026","repository_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/dcondrey%2Fsimpletext-clef2026","tags_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/dcondrey%2Fsimpletext-clef2026/tags","releases_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/dcondrey%2Fsimpletext-clef2026/releases","manifests_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/dcondrey%2Fsimpletext-clef2026/manifests","owner_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners/dcondrey","download_url":"https://codeload.github.com/dcondrey/simpletext-clef2026/tar.gz/refs/heads/main","sbom_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/dcondrey%2Fsimpletext-clef2026/sbom","scorecard":null,"host":{"name":"GitHub","url":"https://github.com","kind":"github","repositories_count":286080680,"owners_count":35245324,"icon_url":"https://github.com/github.png","version":null,"created_at":"2022-05-30T11:31:42.601Z","updated_at":"2026-05-26T15:22:16.424Z","status":"online","status_checked_at":"2026-07-07T02:00:07.222Z","response_time":90,"last_error":null,"robots_txt_status":"success","robots_txt_updated_at":"2025-07-24T06:49:26.215Z","robots_txt_url":"https://github.com/robots.txt","online":true,"can_crawl_api":true,"host_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub","repositories_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories","repository_names_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repository_names","owners_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners"}},"keywords":["clef2026","deberta","hallucination-detection","lightgbm","llm","nlp","simpletext","text-simplification"],"created_at":"2026-07-07T23:01:25.533Z","updated_at":"2026-07-07T23:01:26.286Z","avatar_url":"https://github.com/dcondrey.png","language":"Python","funding_links":[],"categories":[],"sub_categories":[],"readme":"# Writerslogic at the CLEF 2026 SimpleText Track\n\n[![Python 3.10+](https://img.shields.io/badge/python-3.10+-blue.svg)](https://www.python.org/downloads/)\n[![CLEF 2026 SimpleText](https://img.shields.io/badge/CLEF%202026-SimpleText-orange.svg)](http://simpletext-project.com/2026/)\n\nTeam `writerslogic` submission to the **CLEF 2026 SimpleText** track — Task 1 (scientific text simplification) and Task 2 (identify and avoid hallucination). Evaluated on English and multilingual biomedical text from Cochrane systematic reviews.\n\n## Official Results\n\n| Task | Metric | Score | Standing |\n|---|---|---|---|\n| **1.1** — sentence simplification | SARI / BLEU | **47.43** / 14.21 | best **sentence-level** system (3rd on the combined Task 1 board, behind two document-level systems) |\n| **2.1** — binary overgeneration ID | doc-level macro F1 | **0.8085** | **2nd** of teams (behind AIIR Lab 0.8197); our DeBERTa identification run (0.8081) tops the identification track |\n| **2.2** — multi-class error classification | accuracy | **0.804** | **2nd** among unique teams (behind AIIR Lab 0.827) |\n\n## Task 1 — Multi-Candidate Simplification (`task1/`)\n\nFor each input sentence we generate **five candidates at varying temperatures** and select the best with a **reference-free SARI-proxy** score that rewards compression, source-word retention, Cochrane Plain-Language-Summary vocabulary, and lexical simplicity. Our best submission uses **Claude Sonnet 4** as the generator (GPT-4o-mini was used during development); a local Qwen3-8B variant trades quality for fully-local inference. The exact prompt and few-shot examples are in the paper.\n\n## Task 2 — Overgeneration / Hallucination Detection (`task2/`)\n\nWe fine-tune **DeBERTa-v3-large (NLI)** on 350K `(source, sentence)` pairs, framing overgeneration detection as natural language inference (source = premise, candidate = hypothesis). This is complemented by an interpretable **LightGBM stacking ensemble** over 70+ features — source-alignment, prompt-leakage patterns, text-quality, repetition, and **sequential \"CRF-like\" features** that capture how overgeneration errors propagate once they begin within a document — with per-class threshold tuning for the imbalanced multi-class task.\n\n## Repository Structure\n\n```\ntask1/   # multi-candidate generation + SARI-proxy reranking\ntask2/   # DeBERTa-v3 NLI classifier + LightGBM stacking ensemble\n```\n\n## Citation\n\n```bibtex\n@inproceedings{condrey2026simpletext,\n  title     = {Writerslogic at the {CLEF} 2026 {SimpleText} Track:\n               Multi-Candidate {LLM} Simplification and Stacked Complexity Spotting},\n  author    = {Condrey, David},\n  booktitle = {Working Notes of CLEF 2026 -- Conference and Labs of the Evaluation Forum},\n  series    = {CEUR Workshop Proceedings},\n  year      = {2026},\n  publisher = {CEUR-WS.org},\n  note      = {to appear}\n}\n```\n","project_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fdcondrey%2Fsimpletext-clef2026","html_url":"https://awesome.ecosyste.ms/projects/github.com%2Fdcondrey%2Fsimpletext-clef2026","lists_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fdcondrey%2Fsimpletext-clef2026/lists"}