We benchmark the performance of segment-level metrics submitted to WMT 2023 using the ACES Challenge Set (Amrhein et al., 2022). The challenge set consists of 36K examples representing challenges from 68 phenomena and covering 146 language pairs. The phenomena range from simple perturbations at the word/character level to more complex errors based on discourse and real-world knowledge. For each metric, we provide a detailed profile of performance over a range of error categories as well as an overall ACES-Score for quick comparison. We also measure the incremental performance of the metrics submitted to both WMT 2023 and 2022. We find that 1) there is no clear winner among the metrics submitted to WMT 2023, and 2) performance change between the 2023 and 2022 versions of the metrics is highly variable. Our recommendations are similar to those from WMT 2022. Metric developers should focus on: building ensembles of metrics from different design families, developing metrics that pay more attention to the source and rely less on surface-level overlap, and carefully determining the influence of multilingual embeddings on MT evaluation.

For sensible progress in natural language processing, it is important that we are aware of the limitations of the evaluation metrics we use. In this work, we evaluate how robust metrics are to non-standardized dialects, i.e. spelling differences in language varieties that do not have a standard orthography. To investigate this, we collect a dataset of human translations and human judgments for automatic machine translations from English to two Swiss German dialects. We further create a challenge set for dialect variation and benchmark existing metrics’ performances. Our results show that existing metrics cannot reliably evaluate Swiss German text generation outputs, especially on segment level. We propose initial design adaptations that increase robustness in the face of non-standardized dialects, although there remains much room for further improvement. The dataset, code, and models are available here: https://github.com/textshuttle/dialect_eval

Most research on hate speech detection has focused on English where a sizeable amount of labeled training data is available. However, to expand hate speech detection into more languages, approaches that require minimal training data are needed. In this paper, we test whether natural language inference (NLI) models which perform well in zero- and few-shot settings can benefit hate speech detection performance in scenarios where only a limited amount of labeled data is available in the target language. Our evaluation on five languages demonstrates large performance improvements of NLI fine-tuning over direct fine-tuning in the target language. However, the effectiveness of previous work that proposed intermediate fine-tuning on English data is hard to match. Only in settings where the English training data does not match the test domain, can our customised NLI-formulation outperform intermediate fine-tuning on English. Based on our extensive experiments, we propose a set of recommendations for hate speech detection in languages where minimal labeled training data is available.

Natural language generation models reproduce and often amplify the biases present in their training data. Previous research explored using sequence-to-sequence rewriting models to transform biased model outputs (or original texts) into more gender-fair language by creating pseudo training data through linguistic rules. However, this approach is not practical for languages with more complex morphology than English. We hypothesise that creating training data in the reverse direction, i.e. starting from gender-fair text, is easier for morphologically complex languages and show that it matches the performance of state-of-the-art rewriting models for English. To eliminate the rule-based nature of data creation, we instead propose using machine translation models to create gender-biased text from real gender-fair text via round-trip translation. Our approach allows us to train a rewriting model for German without the need for elaborate handcrafted rules. The outputs of this model increased gender-fairness as shown in a human evaluation study.

Neural metrics have achieved impressive correlation with human judgements in the evaluation of machine translation systems, but before we can safely optimise towards such metrics, we should be aware of (and ideally eliminate) biases toward bad translations that receive high scores. Our experiments show that sample-based Minimum Bayes Risk decoding can be used to explore and quantify such weaknesses. When applying this strategy to COMET for en-de and de-en, we find that COMET models are not sensitive enough to discrepancies in numbers and named entities. We further show that these biases are hard to fully remove by simply training on additional synthetic data and release our code and data for facilitating further experiments.

For real-life applications, it is crucial that end-to-end spoken language translation models perform well on continuous audio, without relying on human-supplied segmentation. For online spoken language translation, where models need to start translating before the full utterance is spoken,most previous work has ignored the segmentation problem. In this paper, we compare various methods for improving models’ robustness towards segmentation errors and different segmentation strategies in both offline and online settings and report results on translation quality, flicker and delay. Our findings on five different language pairs show that a simple fixed-window audio segmentation can perform surprisingly well given the right conditions.

As machine translation (MT) metrics improve their correlation with human judgement every year, it is crucial to understand the limitations of these 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.

Many sequence-to-sequence tasks in natural language processing are roughly monotonic in the alignment between source and target sequence, and previous work has facilitated or enforced learning of monotonic attention behavior via specialized attention functions or pretraining. In this work, we introduce a monotonicity loss function that is compatible with standard attention mechanisms and test it on several sequence-to-sequence tasks: grapheme-to-phoneme conversion, morphological inflection, transliteration, and dialect normalization. Experiments show that we can achieve largely monotonic behavior. Performance is mixed, with larger gains on top of RNN baselines. General monotonicity does not benefit transformer multihead attention, however, we see isolated improvements when only a subset of heads is biased towards monotonic behavior.

Data-driven subword segmentation has become the default strategy for open-vocabulary machine translation and other NLP tasks, but may not be sufficiently generic for optimal learning of non-concatenative morphology. We design a test suite to evaluate segmentation strategies on different types of morphological phenomena in a controlled, semi-synthetic setting. In our experiments, we compare how well machine translation models trained on subword- and character-level can translate these morphological phenomena. We find that learning to analyse and generate morphologically complex surface representations is still challenging, especially for non-concatenative morphological phenomena like reduplication or vowel harmony and for rare word stems. Based on our results, we recommend that novel text representation strategies be tested on a range of typologically diverse languages to minimise the risk of adopting a strategy that inadvertently disadvantages certain languages.

Transfer learning is a popular strategy to improve the quality of low-resource machine translation. For an optimal transfer of the embedding layer, the child and parent model should share a substantial part of the vocabulary. This is not the case when transferring to languages with a different script. We explore the benefit of romanization in this scenario. Our results show that romanization entails information loss and is thus not always superior to simpler vocabulary transfer methods, but can improve the transfer between related languages with different scripts. We compare two romanization tools and find that they exhibit different degrees of information loss, which affects translation quality. Finally, we extend romanization to the target side, showing that this can be a successful strategy when coupled with a simple deromanization model.