ACES: Translation Accuracy Challenge Sets for Evaluating Machine Translation Metrics

27 Oct 2022  ยท  Chantal Amrhein, Nikita Moghe, Liane Guillou ยท

As machine translation (MT) metrics improve their correlation with human judgement every year, it is crucial to understand the limitations of such metrics at the segment level. Specifically, it is important to investigate metric behaviour when facing accuracy errors in MT because these can have dangerous consequences in certain contexts (e.g., legal, medical). We curate ACES, a translation accuracy challenge set, consisting of 68 phenomena ranging from simple perturbations at the word/character level to more complex errors based on discourse and real-world knowledge. We use ACES to evaluate a wide range of MT metrics including the submissions to the WMT 2022 metrics shared task and perform several analyses leading to general recommendations for metric developers. We recommend: a) combining metrics with different strengths, b) developing metrics that give more weight to the source and less to surface-level overlap with the reference and c) explicitly modelling additional language-specific information beyond what is available via multilingual embeddings.

PDF Abstract

Datasets


Introduced in the Paper:

ACES

Used in the Paper:

XNLI PAWS-X FLoRes-101 FLoRes-200 Wino-X

Results from the Paper


Task Dataset Model Metric Name Metric Value Global Rank Result Benchmark
Machine Translation ACES HWTSC-Teacher-Sim Score 19.97 # 1
Machine Translation ACES UniTE-src Score 15.68 # 8
Machine Translation ACES MS-COMET-QE-22 Score 19.76 # 3
Machine Translation ACES KG-BERTScore Score 17.28 # 4
Machine Translation ACES Cross-QE Score 14.07 # 12
Machine Translation ACES UniTE-ref Score 15.38 # 9
Machine Translation ACES UniTE Score 14.76 # 11
Machine Translation ACES MS-COMET-22 Score 19.89 # 2
Machine Translation ACES metricx_xxl_DA_2019 Score 15.24 # 10
Machine Translation ACES metricx_xl_MQM_2020 Score 13.08 # 14
Machine Translation ACES metricx_xl_DA_2019 Score 17.17 # 5
Machine Translation ACES COMET-22 Score 16.31 # 7
Machine Translation ACES YiSi-1 Score 11.38 # 17
Machine Translation ACES COMET-QE Score 16.8 # 6
Machine Translation ACES COMET-20 Score 12.06 # 15
Machine Translation ACES BLEURT-20 Score 11.9 # 16
Machine Translation ACES BERTScore Score 10.47 # 18
Machine Translation ACES chrF Score 13.57 # 13
Machine Translation ACES f200spBLEU Score -0.18 # 19
Machine Translation ACES f101spBLEU Score -0.33 # 19
Machine Translation ACES BLEU Score -3.13 # 19

Methods


No methods listed for this paper. Add relevant methods here