| --- |
| title: README |
| emoji: ☄️ |
| colorFrom: red |
| colorTo: blue |
| sdk: static |
| pinned: false |
| --- |
| |
| <a href="https://huggingface.co/spaces/TurCOMET/paper"><img src="https://huggingface.co/spaces/TurCOMET/README/resolve/main/banner.png" alt="TurCOMET: a morphologically-aware evaluation system for document-level Turkish-English machine translation. EMNLP 2026."></a> |
|
|
| # TurCOMET: A Morphologically-Aware Evaluation System for Document-Level Turkish-English Machine Translation |
|
|
| **Atahan Uz · Hüseyin Emir Akdağ · Susan Üsküdarlı**<br> |
| Boğaziçi University · *EMNLP 2026, Budapest, Hungary* |
|
|
| [📄 **Paper (PDF)**](https://huggingface.co/spaces/TurCOMET/paper/resolve/main/TurCOMET_EMNLP2026_paper.pdf) · |
| [🖼️ **Poster (PDF)**](https://huggingface.co/spaces/TurCOMET/paper/resolve/main/TurCOMET_EMNLP2026_poster.pdf) · |
| [🌐 **Project page**](https://huggingface.co/spaces/TurCOMET/paper) · |
| [🤖 **Model**](https://huggingface.co/TurCOMET/XCOMET-XL-TR) · |
| [📚 **TurkDoc-MT**](https://huggingface.co/datasets/TurCOMET/TurkDoc-MT) |
|
|
| TurCOMET is the first Turkish-specific augmentation of COMET. It adds a 330K-parameter bottleneck to a frozen xCOMET-XL that fuses five Turkish-aware lexical features (chrF++, lemma-TER via Zemberek, LaBSE cosine, length-z and a direction flag) into the pooled representation. This counters the sub-word over-segmentation that makes neural metrics miss errors in agglutinative Turkish. On **TurkDoc-MT**, our new 1,322-document, 7-domain EN↔TR benchmark, an expert MQM audit of 11 LLMs shows that TurCOMET agrees with professional translators better than both BLEU and xCOMET-XL. |
|
|
| ## 📖 Read the paper & poster |
|
|
| | Full paper · 21 pages | EMNLP 2026 poster · A0 | |
| |:---:|:---:| |
| | <a href="https://huggingface.co/spaces/TurCOMET/paper#paper"><img src="https://huggingface.co/spaces/TurCOMET/README/resolve/main/paper.webp" width="300" alt="First page of the TurCOMET paper"></a> | <a href="https://huggingface.co/spaces/TurCOMET/paper#poster"><img src="https://huggingface.co/spaces/TurCOMET/README/resolve/main/poster.webp" width="300" alt="TurCOMET EMNLP 2026 poster"></a> | |
| | [Read online](https://huggingface.co/spaces/TurCOMET/paper#paper) · [PDF](https://huggingface.co/spaces/TurCOMET/paper/resolve/main/TurCOMET_EMNLP2026_paper.pdf) | [View online](https://huggingface.co/spaces/TurCOMET/paper#poster) · [PDF](https://huggingface.co/spaces/TurCOMET/paper/resolve/main/TurCOMET_EMNLP2026_poster.pdf) | |
|
|
| ## 📊 Key results |
|
|
| Agreement with expert MQM judgments from 4 professional translators (550 rated translations, 22 model × direction systems): |
|
|
| | Metric | Segment Kendall τ | System Kendall τ | System Pearson r | |
| |:---|---:|---:|---:| |
| | BLEU | 0.256 | 0.628 | 0.828 | |
| | xCOMET-XL | 0.267 | 0.636 | 0.950 | |
| | **TurCOMET (ours)** | **0.291** | **0.749** | **0.976** | |
|
|
| The segment-level gain is significant overall (Δτ = +0.0247, p = 0.022) and comes almost entirely from TR→EN (Δτ = +0.049, p < 0.001). In EN→TR, TurCOMET and xCOMET-XL are statistically tied. |
|
|
| ## 🤗 Released resources |
|
|
| | | Repository | Contents | |
| |:---|:---|:---| |
| | 🤖 Model | [`TurCOMET/XCOMET-XL-TR`](https://huggingface.co/TurCOMET/XCOMET-XL-TR) | TurCOMET checkpoint (bf16, ~7 GB) for scoring EN↔TR translations | |
| | 📚 Benchmark | [`TurCOMET/TurkDoc-MT`](https://huggingface.co/datasets/TurCOMET/TurkDoc-MT) | 1,322 document-level EN↔TR pairs across 7 domains | |
| | ✍️ Annotations | [`TurCOMET/TurkDoc-MT-MQM-Annotations`](https://huggingface.co/datasets/TurCOMET/TurkDoc-MT-MQM-Annotations) | 2,200 expert MQM annotations with error spans, categories and severities | |
| | 🔁 Outputs | [`TurCOMET/TurkDoc-MT_with_model-translations`](https://huggingface.co/datasets/TurCOMET/TurkDoc-MT_with_model-translations) | TurkDoc-MT translations from all 11 evaluated LLMs, both directions | |
| | 🏋️ Training data | [`TurCOMET/wmt-da-en-tr_tr-en`](https://huggingface.co/datasets/TurCOMET/wmt-da-en-tr_tr-en) | 35,357 WMT EN↔TR DA ratings with features and the exact 90/5/5 splits | |
|
|
| ## 📝 Citation |
|
|
| ```bibtex |
| @inproceedings{uz-etal-2026-turcomet, |
| title = {{TurCOMET}: A Morphologically-Aware Evaluation System for |
| Document-Level {T}urkish-{E}nglish Machine Translation}, |
| author = {Uz, Atahan and Akda{\u{g}}, H{\"u}seyin Emir and |
| {\"U}sk{\"u}darl{\i}, Susan}, |
| booktitle = {Proceedings of the 2026 Conference on Empirical Methods |
| in Natural Language Processing}, |
| year = {2026}, |
| address = {Budapest, Hungary}, |
| publisher = {Association for Computational Linguistics} |
| } |
| ``` |
|
|
