Unsupervised Word-level Quality Estimation for Machine Translation Through the Lens of Annotators (Dis)agreement
Fuente:
arXiv
Salvato in:
| Autori principali: | , , , |
|---|---|
| Natura: | Preprint |
| Pubblicazione: |
2025
|
| Soggetti: | |
| Accesso online: | |
| Tags: |
Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
|
| _version_ | 1866915619592470528 |
|---|---|
| author | Sarti, Gabriele Zouhar, Vilém Nissim, Malvina Bisazza, Arianna |
| author_facet | Sarti, Gabriele Zouhar, Vilém Nissim, Malvina Bisazza, Arianna |
| contents | Word-level quality estimation (WQE) aims to automatically identify fine-grained error spans in machine-translated outputs and has found many uses, including assisting translators during post-editing. Modern WQE techniques are often expensive, involving prompting of large language models or ad-hoc training on large amounts of human-labeled data. In this work, we investigate efficient alternatives exploiting recent advances in language model interpretability and uncertainty quantification to identify translation errors from the inner workings of translation models. In our evaluation spanning 14 metrics across 12 translation directions, we quantify the impact of human label variation on metric performance by using multiple sets of human labels. Our results highlight the untapped potential of unsupervised metrics, the shortcomings of supervised methods when faced with label uncertainty, and the brittleness of single-annotator evaluation practices. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2505_23183 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Unsupervised Word-level Quality Estimation for Machine Translation Through the Lens of Annotators (Dis)agreement Sarti, Gabriele Zouhar, Vilém Nissim, Malvina Bisazza, Arianna Computation and Language Artificial Intelligence Human-Computer Interaction Word-level quality estimation (WQE) aims to automatically identify fine-grained error spans in machine-translated outputs and has found many uses, including assisting translators during post-editing. Modern WQE techniques are often expensive, involving prompting of large language models or ad-hoc training on large amounts of human-labeled data. In this work, we investigate efficient alternatives exploiting recent advances in language model interpretability and uncertainty quantification to identify translation errors from the inner workings of translation models. In our evaluation spanning 14 metrics across 12 translation directions, we quantify the impact of human label variation on metric performance by using multiple sets of human labels. Our results highlight the untapped potential of unsupervised metrics, the shortcomings of supervised methods when faced with label uncertainty, and the brittleness of single-annotator evaluation practices. |
| title | Unsupervised Word-level Quality Estimation for Machine Translation Through the Lens of Annotators (Dis)agreement |
| topic | Computation and Language Artificial Intelligence Human-Computer Interaction |
| url | https://arxiv.org/abs/2505.23183 |