Recent research in neural machine translation (NMT) has shown that training on high-quality machine-generated data can outperform training on human-generated data. This work accompanies the first-ever release of a LLM-generated, MBR-decoded and QE-reranked dataset with both sentence-level and multi-sentence examples. We perform extensive experiments to demonstrate the quality of our dataset in terms of its downstream impact on NMT model performance. We find that training from scratch on our (machine-generated) dataset outperforms training on the (web-crawled) WMT’23 training dataset (which is 300 times larger), and also outperforms training on the top-quality subset of the WMT’23 training dataset. We also find that performing self-distillation by finetuning the LLM which generated this dataset outperforms the LLM’s strong few-shot baseline. These findings corroborate the quality of our dataset, and demonstrate the value of high-quality machine-generated data in improving performance of NMT models.

Maximum-a-posteriori (MAP) decoding is the most widely used decoding strategy for neural machine translation (NMT) models. The underlying assumption is that model probability correlates well with human judgment, with better translations getting assigned a higher score by the model. However, research has shown that this assumption does not always hold, and generation quality can be improved by decoding to optimize a utility function backed by a metric or quality-estimation signal, as is done by Minimum Bayes Risk (MBR) or Quality-Aware decoding. The main disadvantage of these approaches is that they require an additional model to calculate the utility function during decoding, significantly increasing the computational cost. In this paper, we propose to make the NMT models themselves quality-aware by training them to estimate the quality of their own output. Using this approach for MBR decoding we can drastically reduce the size of the candidate list, resulting in a speed-up of two-orders of magnitude. When applying our method to MAP decoding we obtain quality gains similar or even superior to quality reranking approaches, but with the efficiency of single pass decoding.

Large language models (LLMs) that have been trained on multilingual but not parallel text exhibit a remarkable ability to translate between languages. We probe this ability in an in-depth study of the pathways language model (PaLM), which has demonstrated the strongest machine translation (MT) performance among similarly-trained LLMs to date. We investigate various strategies for choosing translation examples for few-shot prompting, concluding that example quality is the most important factor. Using optimized prompts, we revisit previous assessments of PaLM’s MT capabilities with more recent test sets, modern MT metrics, and human evaluation, and find that its performance, while impressive, still lags that of state-of-the-art supervised systems. We conclude by providing an analysis of PaLM’s MT output which reveals some interesting properties and prospects for future work.

Quality Estimation (QE), the evaluation of machine translation output without the need of explicit references, has seen big improvements in the last years with the use of neural metrics. In this paper we analyze the viability of using QE metrics for filtering out bad quality sentence pairs in the training data of neural machine translation systems (NMT). While most corpus filtering methods are focused on detecting noisy examples in collections of texts, usually huge amounts of web crawled data, QE models are trained to discriminate more fine-grained quality differences. We show that by selecting the highest quality sentence pairs in the training data, we can improve translation quality while reducing the training size by half. We also provide a detailed analysis of the filtering results, which highlights the differences between both approaches.

Machine translation (MT) evaluation often focuses on accuracy and fluency, without paying much attention to translation style. This means that, even when considered accurate and fluent, MT output can still sound less natural than high quality human translations or text originally written in the target language. Machine translation output notably exhibits lower lexical diversity, and employs constructs that mirror those in the source sentence. In this work we propose a method for training MT systems to achieve a more natural style, i.e. mirroring the style of text originally written in the target language. Our method tags parallel training data according to the naturalness of the target side by contrasting language models trained on natural and translated data. Tagging data allows us to put greater emphasis on target sentences originally written in the target language. Automatic metrics show that the resulting models achieve lexical richness on par with human translations, mimicking a style much closer to sentences originally written in the target language. Furthermore, we find that their output is preferred by human experts when compared to the baseline translations.

Sub-word segmentation is currently a standard tool for training neural machine translation (MT) systems and other NLP tasks. The goal is to split words (both in the source and target languages) into smaller units which then constitute the input and output vocabularies of the MT system. The aim of reducing the size of the input and output vocabularies is to increase the generalization capabilities of the translation model, enabling the system to translate and generate infrequent and new (unseen) words at inference time by combining previously seen sub-word units. Ideally, we would expect the created units to have some linguistic meaning, so that words are created in a compositional way. However, the most popular word-splitting method, Byte-Pair Encoding (BPE), which originates from the data compression literature, does not include explicit criteria to favor linguistic splittings nor to find the optimal sub-word granularity for the given training data. In this paper, we propose a statistically motivated extension of the BPE algorithm and an effective convergence criterion that avoids the costly experimentation cycle needed to select the best sub-word vocabulary size. Experimental results with morphologically rich languages show that our model achieves nearly-optimal BLEU scores and produces morphologically better word segmentations, which allows to outperform BPE’s generalization in the translation of sentences containing new words, as shown via human evaluation.

Training data for machine translation (MT) is often sourced from a multitude of large corpora that are multi-faceted in nature, e.g. containing contents from multiple domains or different levels of quality or complexity. Naturally, these facets do not occur with equal frequency, nor are they equally important for the test scenario at hand. In this work, we propose to optimize this balance jointly with MT model parameters to relieve system developers from manual schedule design. A multi-armed bandit is trained to dynamically choose between facets in a way that is most beneficial for the MT system. We evaluate it on three different multi-facet applications: balancing translationese and natural training data, or data from multiple domains or multiple language pairs. We find that bandit learning leads to competitive MT systems across tasks, and our analysis provides insights into its learned strategies and the underlying data sets.

Fine-grained control of machine translation (MT) outputs along multiple attributes is critical for many modern MT applications and is a requirement for gaining users’ trust. A standard approach for exerting control in MT is to prepend the input with a special tag to signal the desired output attribute. Despite its simplicity, attribute tagging has several drawbacks: continuous values must be binned into discrete categories, which is unnatural for certain applications; interference between multiple tags is poorly understood. We address these problems by introducing vector-valued interventions which allow for fine-grained control over multiple attributes simultaneously via a weighted linear combination of the corresponding vectors. For some attributes, our approach even allows for fine-tuning a model trained without annotations to support such interventions. In experiments with three attributes (length, politeness and monotonicity) and two language pairs (English to German and Japanese) our models achieve better control over a wider range of tasks compared to tagging, and translation quality does not degrade when no control is requested. Finally, we demonstrate how to enable control in an already trained model after a relatively cheap fine-tuning stage.

We present Sockeye 2, a modernized and streamlined version of the Sockeye neural machine translation (NMT) toolkit. New features include a simplified code base through the use of MXNet’s Gluon API, a focus on state of the art model architectures, and distributed mixed precision training. These improvements result in faster training and inference, higher automatic metric scores, and a shorter path from research to production.

The end-to-end nature of neural machine translation (NMT) removes many ways of manually guiding the translation process that were available in older paradigms. Recent work, however, has introduced a new capability: lexically constrained or guided decoding, a modification to beam search that forces the inclusion of pre-specified words and phrases in the output. However, while theoretically sound, existing approaches have computational complexities that are either linear (Hokamp and Liu, 2017) or exponential (Anderson et al., 2017) in the number of constraints. We present a algorithm for lexically constrained decoding with a complexity of O(1) in the number of constraints. We demonstrate the algorithm’s remarkable ability to properly place these constraints, and use it to explore the shaky relationship between model and BLEU scores. Our implementation is available as part of Sockeye.

In this paper we explore the use of Learning Hidden Unit Contribution for the task of neural machine translation. The method was initially proposed in the context of speech recognition for adapting a general system to the specific acoustic characteristics of each speaker. Similar in spirit, in a machine translation framework we want to adapt a general system to a specific domain. We show that the proposed method achieves improvements of up to 2.6 BLEU points over a general system, and up to 6 BLEU points if the initial system has only been trained on out-of-domain data, a situation which may easily happen in practice. The good performance together with its short training time and small memory footprint make it a very attractive solution for domain adaptation.

Human translators are the key to evaluating machine translation (MT) quality and also to addressing the so far unanswered question when and how to use MT in professional translation workflows. This paper describes the corpus developed as a result of a detailed large scale human evaluation consisting of three tightly connected tasks: ranking, error classification and post-editing.

Significant breakthroughs in machine translation only seem possible if human translators are taken into the loop. While automatic evaluation and scoring mechanisms such as BLEU have enabled the fast development of systems, it is not clear how systems can meet real-world (quality) requirements in industrial translation scenarios today. The taraXÜ project paves the way for wide usage of hybrid machine translation outputs through various feedback loops in system development. In a consortium of research and industry partners, the project integrates human translators into the development process for rating and post-editing of machine translation outputs thus collecting feedback for possible improvements.

We describe DFKI’s submission to the System Combination and Machine Translation tracks of the 2011 IWSLT Evaluation Campaign. We focus on a sentence selection mechanism which chooses the (hopefully) best sentence among a set of candidates. The rationale behind it is to take advantage of the strengths of each system, especially given an heterogeneous dataset like the one in this evaluation campaign, composed of TED Talks of very different topics. We focus on using features that correlate well with human judgement and, while our primary system still focus on optimizing the BLEU score on the development set, our goal is to move towards optimizing directly the correlation with human judgement. This kind of system is still under development and was used as a secondary submission.

In this work we review and compare three additional syntactic enhancements for the hierarchical phrase-based translation model, which have been presented in the last few years. We compare their performance when applied separately and study whether the combination may yield additional improvements. Our findings show that the models are complementary, and their combination achieve an increase of 1% in BLEU and a reduction of nearly 2% in TER. The models presented in this work are made available as part of the Jane open source machine translation toolkit.

In this paper we describe the statistical machine translation system of the RWTH Aachen University developed for the translation task of the IWSLT 2010. This year, we participated in the BTEC translation task for the Arabic to English language direction. We experimented with two state-of-theart decoders: phrase-based and hierarchical-based decoders. Extensions to the decoders included phrase training (as opposed to heuristic phrase extraction) for the phrase-based decoder, and soft syntactic features for the hierarchical decoder. Additionally, we experimented with various rule-based and statistical-based segmenters for Arabic. Due to the different decoders and the different methodologies that we apply for segmentation, we expect that there will be complimentary variation in the results achieved by each system. The next step would be to exploit these variations and achieve better results by combining the systems. We try different strategies for system combination and report significant improvements over the best single system.

Currently most state-of-the-art statistical machine translation systems present a mismatch between training and generation conditions. Word alignments are computed using the well known IBM models for single-word based translation. Afterwards phrases are extracted using extraction heuristics, unrelated to the stochastic models applied for finding the word alignment. In the last years, several research groups have tried to overcome this mismatch, but only with limited success. Recently, the technique of forced alignments has shown to improve translation quality for a phrase-based system, applying a more statistically sound approach to phrase extraction. In this work we investigate the first steps to combine forced alignment with a hierarchical model. Experimental results on IWSLT and WMT data show improvements in translation quality of up to 0.7% BLEU and 1.0% TER.

RWTH’s system for the 2008 IWSLT evaluation consists of a combination of different phrase-based and hierarchical statistical machine translation systems. We participated in the translation tasks for the Chinese-to-English and Arabic-to-English language pairs. We investigated different preprocessing techniques, reordering methods for the phrase-based system, including reordering of speech lattices, and syntax-based enhancements for the hierarchical systems. We also tried the combination of the Arabic-to-English and Chinese-to-English outputs as an additional submission.

Similar to phrase-based machine translation, hierarchical systems produce a large proportion of phrases, most of which are supposedly junk and useless for the actual translation. For the hierarchical case, however, the amount of extracted rules is an order of magnitude bigger. In this paper, we investigate several soft constraints in the extraction of hierarchical phrases and whether these help as additional scores in the decoding to prune unneeded phrases. We show the methods that help best.

The RWTH system for the IWSLT 2007 evaluation is a combination of several statistical machine translation systems. The combination includes Phrase-Based models, a n-gram translation model and a hierarchical phrase model. We describe the individual systems and the method that was used for combining the system outputs. Compared to our 2006 system, we newly introduce a hierarchical phrase-based translation model and show improvements in system combination for Machine Translation. RWTH participated in the Italian-to-English and Chinese-to-English translation directions.

Evaluation of automatic translation output is a difficult task. Several performance measures like Word Error Rate, Position Independent Word Error Rate and the BLEU and NIST scores are widely use and provide a useful tool for comparing different systems and to evaluate improvements within a system. However the interpretation of all of these measures is not at all clear, and the identification of the most prominent source of errors in a given system using these measures alone is not possible. Therefore some analysis of the generated translations is needed in order to identify the main problems and to focus the research efforts. This area is however mostly unexplored and few works have dealt with it until now. In this paper we will present a framework for classification of the errors of a machine translation system and we will carry out an error analysis of the system used by the RWTH in the first TC-STAR evaluation.

In this paper we present the ongoing work at RWTH Aachen University for building a speech-to-speech translation system within the TC-Star project. The corpus we work on consists of parliamentary speeches held in the European Plenary Sessions. To our knowledge, this is the first project that focuses on speech-to-speech translation applied to a real-life task. We describe the statistical approach used in the development of our system and analyze its performance under different conditions: dealing with syntactically correct input, dealing with the exact transcription of speech and dealing with the (noisy) output of an automatic speech recognition system. Experimental results show that our system is able to perform adequately in each of these conditions.