This article investigates the performance of automatic evaluation metrics (AEMs) and LLM-as-a-judge evaluation on literary translation across multiple languages, genres, and translation modalities. The aim is to assess how well these tools align with professionals when evaluating translation, creativity (creative shifts & errors), and see if they can substitute laborious manual annotations. A dataset of literary translations across three modalities (human translation, machine translation, and post-editing), three genres and three language pairs was created and annotated in detail for creativity by experienced professional literary translators. The results show that both AEMs and LLM-as-a-judge evaluations correlate poorly with professional evaluations on creativity, with LLM-as-a-judge showing a systematic bias in favour of machine-translated texts and penalising creative and culturally appropriate solutions. Moreover, performance is consistently worse for more literary genres such as poetry. This highlights fundamental limitations of current automatic evaluation tools for literary translation and the need to create new tools that do not frequently consider out of routine translations as errors.

Neural machine translation (NMT) systems amplify lexical biases present in their training data, leading to artificially impoverished language in output translations.These language-level characteristics render automatic translations different from text originally written in a language and human translations, which hinders their usefulness in for example creating evaluation datasets.Attempts to increase naturalness in NMT can fall short in terms of content preservation, where increased lexical diversity comes at the cost of translation accuracy.Inspired by the reinforcement learning from human feedback framework, we introduce a novel method that rewards both naturalness and content preservation.We experiment with multiple perspectives to produce more natural translations, aiming at reducing machine and human translationese.We evaluate our method on English-to-Dutch literary translation, and find that our best model produces translations that are lexically richer and exhibit more properties of human-written language, without loss in translation accuracy.

Parallel corpora play a vital role in advanced multilingual natural language processing tasks, notably in machine translation (MT). The recent emergence of numerous large parallel corpora, often extracted from multilingual documents on the Internet, has expanded the available resources. Nevertheless, the quality of these corpora remains largely unexplored, while there are large differences in how the corpora are constructed. Moreover, how the potential differences affect the performance of neural MT (NMT) systems has also received limited attention. This study addresses this gap by manually and automatically evaluating four well-known publicly available parallel corpora across eleven language pairs. Our findings are quite concerning: all corpora contain a substantial amount of noisy sentence pairs, with CCMatrix and CCAligned having well below of 50% reasonably clean pairs. MaCoCu and ParaCrawl generally have higher quality texts, though around a third of the texts still have clear issues. While corpus size impacts NMT models’ performance, our study highlights the critical role of quality: higher-quality corpora consistently yield better-performing NMT models when controlling for size.

Large, curated, web-crawled corpora play a vital role in training language models (LMs). They form the lion’s share of the training data in virtually all recent LMs, such as the well-known GPT, LLaMA and XLM-RoBERTa models. However, despite this importance, relatively little attention has been given to the quality of these corpora. In this paper, we compare four of the currently most relevant large, web-crawled corpora (CC100, MaCoCu, mC4 and OSCAR) across eleven lower-resourced European languages. Our approach is two-fold: first, we perform an intrinsic evaluation by performing a human evaluation of the quality of samples taken from different corpora; then, we assess the practical impact of the qualitative differences by training specific LMs on each of the corpora and evaluating their performance on downstream tasks. We find that there are clear differences in quality of the corpora, with MaCoCu and OSCAR obtaining the best results. However, during the extrinsic evaluation, we actually find that the CC100 corpus achieves the highest scores. We conclude that, in our experiments, the quality of the web-crawled corpora does not seem to play a significant role when training LMs.

Machine translations are found to be lexically poorer than human translations. The loss of lexical diversity through MT poses an issue in the automatic translation of litrature, where it matters not only what is written, but also how it is written. Current methods for increasing lexical diversity in MT are rigid. Yet, as we demonstrate, the degree of lexical diversity can vary considerably across different novels. Thus, rather than aiming for the rigid increase of lexical diversity, we reframe the task as recovering what is lost in the machine translation process. We propose a novel approach that consists of reranking translation candidates with a classifier that distinguishes between original and translated text. We evaluate our approach on 31 English-to-Dutch book translations, and find that, for certain books, our approach retrieves lexical diversity scores that are close to human translation.

Acoustic-to-articulatory inversion (AAI) is the process of inferring vocal tract movements from acoustic speech signals. Despite its diverse potential applications, AAI research in languages other than English is scarce due to the challenges of collecting articulatory data. In recent years, self-supervised learning (SSL) based representations have shown great potential for addressing low-resource tasks. We utilize wav2vec 2.0 representations and English articulatory data for training AAI systems and investigates their effectiveness for a different language: Dutch. Results show that using mms-1b features can reduce the cross-lingual performance drop to less than 30%. We found that increasing model size, selecting intermediate rather than final layers, and including more pre-training data improved AAI performance. By contrast, fine-tuning on an ASR task did not. Our results therefore highlight promising prospects for implementing SSL in AAI for languages with limited articulatory data.

We present the most relevant results of the project MaCoCu: Massive collection and curation of monolingual and bilingual data: focus on under-resourced languages in its second year. To date, parallel and monolingual corpora have been produced for seven low-resourced European languages by crawling large amounts of textual data from selected top-level domains of the Internet; both human and automatic evaluation show its usefulness. In addition, several large language models pretrained on MaCoCu data have been published, as well as the code used to collect and curate the data.

We tackle the task of automatically discriminating between human and machine translations. As opposed to most previous work, we perform experiments in a multilingual setting, considering multiple languages and multilingual pretrained language models. We show that a classifier trained on parallel data with a single source language (in our case German–English) can still perform well on English translations that come from different source languages, even when the machine translations were produced by other systems than the one it was trained on. Additionally, we demonstrate that incorporating the source text in the input of a multilingual classifier improves (i) its accuracy and (ii) its robustness on cross-system evaluation, compared to a monolingual classifier. Furthermore, we find that using training data from multiple source languages (German, Russian and Chinese) tends to improve the accuracy of both monolingual and multilingual classifiers. Finally, we show that bilingual classifiers and classifiers trained on multiple source languages benefit from being trained on longer text sequences, rather than on sentences.

Even though many recent semantic parsers are based on deep learning methods, we should not forget that rule-based alternatives might offer advantages over neural approaches with respect to transparency, portability, and explainability. Taking advantage of existing off-the-shelf Universal Dependency parsers, we present a method that maps a syntactic dependency tree to a formal meaning representation based on Discourse Representation Theory. Rather than using lambda calculus to manage variable bindings, our approach is novel in that it consists of using a series of graph transformations. The resulting UD semantic parser shows good performance for English, German, Italian and Dutch, with F-scores over 75%, outperforming a neural semantic parser for the lower-resourced languages. Unlike neural semantic parsers, our UD semantic parser does not hallucinate output, is relatively easy to port to other languages, and is completely transparent.

We introduce the project “MaCoCu: Massive collection and curation of monolingual and bilingual data: focus on under-resourced languages”, funded by the Connecting Europe Facility, which is aimed at building monolingual and parallel corpora for under-resourced European languages. The approach followed consists of crawling large amounts of textual data from carefully selected top-level domains of the Internet, and then applying a curation and enrichment pipeline. In addition to corpora, the project will release successive versions of the free/open-source web crawling and curation software used.

We address the task of automatically distinguishing between human-translated (HT) and machine translated (MT) texts. Following recent work, we fine-tune pre-trained language models (LMs) to perform this task. Our work differs in that we use state-of-the-art pre-trained LMs, as well as the test sets of the WMT news shared tasks as training data, to ensure the sentences were not seen during training of the MT system itself. Moreover, we analyse performance for a number of different experimental setups, such as adding translationese data, going beyond the sentence-level and normalizing punctuation. We show that (i) choosing a state-of-the-art LM can make quite a difference: our best baseline system (DeBERTa) outperforms both BERT and RoBERTa by over 3% accuracy, (ii) adding translationese data is only beneficial if there is not much data available, (iii) considerable improvements can be obtained by classifying at the document-level and (iv) normalizing punctuation and thus avoiding (some) shortcuts has no impact on model performance.

We combine character-level and contextual language model representations to improve performance on Discourse Representation Structure parsing.Character representations can easily be added in a sequence-to-sequence model in either one encoder or as a fully separate encoder, with improvements that are robust to different language models, languages and data sets.For English, these improvements are larger than adding individual sources of linguistic information or adding non-contextual embeddings.A new method of analysis based on semantic tags demonstrates that the character-level representations improve performance across a subset of selected semantic phenomena.

Analogies such as man is to king as woman is to X are often used to illustrate the amazing power of word embeddings. Concurrently, they have also been used to expose how strongly human biases are encoded in vector spaces trained on natural language, with examples like man is to computer programmer as woman is to homemaker. Recent work has shown that analogies are in fact not an accurate diagnostic for bias, but this does not mean that they are not used anymore, or that their legacy is fading. Instead of focusing on the intrinsic problems of the analogy task as a bias detection tool, we discuss a series of issues involving implementation as well as subjective choices that might have yielded a distorted picture of bias in word embeddings. We stand by the truth that human biases are present in word embeddings, and, of course, the need to address them. But analogies are not an accurate tool to do so, and the way they have been most often used has exacerbated some possibly non-existing biases and perhaps hidden others. Because they are still widely popular, and some of them have become classics within and outside the NLP community, we deem it important to provide a series of clarifications that should put well-known, and potentially new analogies, into the right perspective.

Neural methods have had several recent successes in semantic parsing, though they have yet to face the challenge of producing meaning representations based on formal semantics. We present a sequence-to-sequence neural semantic parser that is able to produce Discourse Representation Structures (DRSs) for English sentences with high accuracy, outperforming traditional DRS parsers. To facilitate the learning of the output, we represent DRSs as a sequence of flat clauses and introduce a method to verify that produced DRSs are well-formed and interpretable. We compare models using characters and words as input and see (somewhat surprisingly) that the former performs better than the latter. We show that eliminating variable names from the output using De Bruijn indices increases parser performance. Adding silver training data boosts performance even further.