# In the Blind: Building Pseudo-References for MT Evaluation

arXiv: https://arxiv.org/abs/2609.13611 · HTML: https://mlsanity.com/paper/2609.13611/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 24 of its 45 references that are in the map.
- First posted on arXiv 11 Sep 2026.

## Builds on (most cited first, 24 of 24)

- [Judging LLM-as-a-Judge with MT-Bench and Chatbot Arena](https://mlsanity.com/paper/2306.05685/) (2023)
- [Enriching Word Vectors with Subword Information](https://mlsanity.com/paper/1607.04606/) (2016)
- [BLEURT: Learning Robust Metrics for Text Generation](https://mlsanity.com/paper/2004.04696/) (2020)
- [COMET: A Neural Framework for MT Evaluation](https://mlsanity.com/paper/2009.09025/) (2020)
- [LLM Evaluators Recognize and Favor Their Own Generations](https://mlsanity.com/paper/2404.13076/) (2024)
- [Large Language Models Are State-of-the-Art Evaluators of Translation Quality](https://mlsanity.com/paper/2302.14520/) (2023)
- [Experts, Errors, and Context: A Large-Scale Study of Human Evaluation for Machine Translation](https://mlsanity.com/paper/2104.14478/) (2021)
- [Improving Massively Multilingual Neural Machine Translation and Zero-Shot Translation](https://mlsanity.com/paper/2004.11867/) (2020)
- [xCOMET: Transparent Machine Translation Evaluation through Fine-grained Error Detection](https://mlsanity.com/paper/2310.10482/) (2023)
- [Tower: An Open Multilingual Large Language Model for Translation-Related Tasks](https://mlsanity.com/paper/2402.17733/) (2024)
- [To Ship or Not to Ship: An Extensive Evaluation of Automatic Metrics for Machine Translation](https://mlsanity.com/paper/2107.10821/) (2021)
- [Is MAP Decoding All You Need? The Inadequacy of the Mode in Neural Machine Translation](https://mlsanity.com/paper/2005.10283/) (2020)
- [BLEU might be Guilty but References are not Innocent](https://mlsanity.com/paper/2004.06063/) (2020)
- [High Quality Rather than High Model Probability: Minimum Bayes Risk Decoding with Neural Metrics](https://mlsanity.com/paper/2111.09388/) (2021)
- [MetricX-24: The Google Submission to the WMT 2024 Metrics Shared Task](https://mlsanity.com/paper/2410.03983/) (2024)
- [Scaling up COMETKIWI: Unbabel-IST 2023 Submission for the Quality Estimation Shared Task](https://mlsanity.com/paper/2309.11925/) (2023)
- [Understanding and Mitigating Language Confusion in LLMs](https://mlsanity.com/paper/2406.20052/) (2024)
- [Error Span Annotation: A Balanced Approach for Human Evaluation of Machine Translation](https://mlsanity.com/paper/2406.11580/) (2024)
- [Leveraging GPT-4 for Automatic Translation Post-Editing](https://mlsanity.com/paper/2305.14878/) (2023)
- [Identifying Weaknesses in Machine Translation Metrics Through Minimum Bayes Risk Decoding: A Case Study for COMET](https://mlsanity.com/paper/2202.05148/) (2022)
- [Tower+: Bridging Generality and Translation Specialization in Multilingual LLMs](https://mlsanity.com/paper/2506.17080/) (2025)
- [BLEURT Has Universal Translations: An Analysis of Automatic Metrics by Minimum Risk Training](https://mlsanity.com/paper/2307.03131/) (2023)
- [TranslateGemma Technical Report](https://mlsanity.com/paper/2601.09012/) (2026)
- [Mending the Holes: Mitigating Reward Hacking in Reinforcement Learning for Multilingual Translation](https://mlsanity.com/paper/2603.13045/) (2026)

## Abstract (arXiv)

The WMT26 General MT task evaluates systems on 10 language pairs that have no human references (neither translated from scratch nor post-edited from MT output by humans). We describe how we built the pseudo-references for these pairs and six other language pairs (in which some forms of human references are available): seven models translate the 3,277 official documents under up to five prompt conditions, giving a total of 26 system-prompt combinations; then three reference-free quality estimation (QE) models score every candidate; and a per-document selector picks one translation, which GPT-5.5 post-edits where needed. Working without references exposed a failure mode of QE-guided selection: the metrics rank fluent output in the wrong language above correct translations. Adding a confidence-scaled language identification penalty to the score fusion drives the wrong-language count to zero, and the resulting selector still scores better on MetricX than the rank-fusion baseline it replaces. Since no references were available for these pairs while we were building them, we calibrate every selection decision on last year's WMT25 human judgments. The human evaluation, released after construction, shows the cost of getting selection wrong: our references stand with the strongest participating systems when the selector kept a frontier-model candidate, and fall up to 17 ESA points below them when it did not. We release the selection method and the provenance of every reference (https://github.com/surrey-nlp/PseudoRef)