Pre-trained language models (PLMs) often take advantage of the monolingual and multilingual dataset that is freely available online to acquire general or mixed domain knowledge before deployment into specific tasks. Extra-large PLMs (xLPLMs) are proposed very recently to claim supreme performances over smaller-sized PLMs such as in machine translation (MT) tasks. These xLPLMs include Meta-AI's wmt21-dense-24-wide-en-X (2021) and NLLB (2022). In this work, we examine if xLPLMs are absolutely superior to smaller-sized PLMs in fine-tuning toward domain-specific MTs. We use two different in-domain data of different sizes: commercial automotive in-house data and clinical shared task data from the ClinSpEn2022 challenge at WMT2022. We choose popular Marian Helsinki as smaller sized PLM and two massive-sized Mega-Transformers from Meta-AI as xLPLMs. Our experimental investigation shows that 1) on smaller-sized in-domain commercial automotive data, xLPLM wmt21-dense-24-wide-en-X indeed shows much better evaluation scores using SacreBLEU and hLEPOR metrics than smaller-sized Marian, even though its score increase rate is lower than Marian after fine-tuning; 2) on relatively larger-size well prepared clinical data fine-tuning, the xLPLM NLLB tends to lose its advantage over smaller-sized Marian on two sub-tasks (clinical terms and ontology concepts) using ClinSpEn offered metrics METEOR, COMET, and ROUGE-L, and totally lost to Marian on Task-1 (clinical cases) on all official metrics including SacreBLEU and BLEU; 3) metrics do not always agree with each other on the same tasks using the same model outputs; 4) clinic-Marian ranked No.2 on Task- 1 (via SACREBLEU/BLEU) and Task-3 (via METEOR and ROUGE) among all submissions.

In NMT we search for the mode of the model distribution to form predictions. The mode and other high-probability translations found by beam search have been shown to often be inadequate in a number of ways. This prevents improving translation quality through better search, as these idiosyncratic translations end up selected by the decoding algorithm, a problem known as the beam search curse. Recently, an approximation to minimum Bayes risk (MBR) decoding has been proposed as an alternative decision rule that would likely not suffer from the same problems. We analyse this approximation and establish that it has no equivalent to the beam search curse. We then design approximations that decouple the cost of exploration from the cost of robust estimation of expected utility. This allows for much larger hypothesis spaces, which we show to be beneficial. We also show that mode-seeking strategies can aid in constructing compact sets of promising hypotheses and that MBR is effective in identifying good translations in them. We conduct experiments on three language pairs varying in amounts of resources available: English into and from German, Romanian, and Nepali.

While machine translation evaluation metrics based on string overlap (e.g., BLEU) have their limitations, their computations are transparent: the BLEU score assigned to a particular candidate translation can be traced back to the presence or absence of certain words. The operations of newer learned metrics (e.g., BLEURT, COMET), which leverage pretrained language models to achieve higher correlations with human quality judgments than BLEU, are opaque in comparison. In this paper, we shed light on the behavior of these learned metrics by creating DEMETR, a diagnostic dataset with 31K English examples (translated from 10 source languages) for evaluating the sensitivity of MT evaluation metrics to 35 different linguistic perturbations spanning semantic, syntactic, and morphological error categories. All perturbations were carefully designed to form minimal pairs with the actual translation (i.e., differ in only one aspect). We find that learned metrics perform substantially better than string-based metrics on DEMETR. Additionally, learned metrics differ in their sensitivity to various phenomena (e.g., BERTScore is sensitive to untranslated words but relatively insensitive to gender manipulation, while COMET is much more sensitive to word repetition than to aspectual changes). We publicly release DEMETR to spur more informed future development of machine translation evaluation metrics

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.

This paper describes Microsoft's submission to the first shared task on sign language translation at WMT 2022, a public competition tackling sign language to spoken language translation for Swiss German sign language. The task is very challenging due to data scarcity and an unprecedented vocabulary size of more than 20k words on the target side. Moreover, the data is taken from real broadcast news, includes native signing and covers scenarios of long videos. Motivated by recent advances in action recognition, we incorporate full body information by extracting features from a pre-trained I3D model and applying a standard transformer network. The accuracy of the system is further improved by applying careful data cleaning on the target text. We obtain BLEU scores of 0.6 and 0.78 on the test and dev set respectively, which is the best score among the participants of the shared task. Also in the human evaluation the submission reaches the first place. The BLEU score is further improved to 1.08 on the dev set by applying features extracted from a lip reading model.

A straightforward approach to context-aware neural machine translation consists in feeding the standard encoder-decoder architecture with a window of consecutive sentences, formed by the current sentence and a number of sentences from its context concatenated to it. In this work, we propose an improved concatenation approach that encourages the model to focus on the translation of the current sentence, discounting the loss generated by target context. We also propose an additional improvement that strengthen the notion of sentence boundaries and of relative sentence distance, facilitating model compliance to the context-discounted objective. We evaluate our approach with both average-translation quality metrics and contrastive test sets for the translation of inter-sentential discourse phenomena, proving its superiority to the vanilla concatenation approach and other sophisticated context-aware systems.

Machine Translation (MT) is usually viewed as a one-shot process that generates the target language equivalent of some source text from scratch. We consider here a more general setting which assumes an initial target sequence, that must be transformed into a valid translation of the source, thereby restoring parallelism between source and target. For this bilingual synchronization task, we consider several architectures (both autoregressive and non-autoregressive) and training regimes, and experiment with multiple practical settings such as simulated interactive MT, translating with Translation Memory (TM) and TM cleaning. Our results suggest that one single generic edit-based system, once fine-tuned, can compare with, or even outperform, dedicated systems specifically trained for these tasks.

Customer feedback can be an important signal for improving commercial machine translation systems. One solution for fixing specific translation errors is to remove the related erroneous training instances followed by re-training of the machine translation system, which we refer to as instance-specific data filtering. Influence functions (IF) have been shown to be effective in finding such relevant training examples for classification tasks such as image classification, toxic speech detection and entailment task. Given a probing instance, IF find influential training examples by measuring the similarity of the probing instance with a set of training examples in gradient space. In this work, we examine the use of influence functions for Neural Machine Translation (NMT). We propose two effective extensions to a state of the art influence function and demonstrate on the sub-problem of copied training examples that IF can be applied more generally than handcrafted regular expressions.

Reference Expression Generation (REG) and Comprehension (REC) are two highly correlated tasks. Modeling REG and REC simultaneously for utilizing the relation between them is a promising way to improve both. However, the problem of distinct inputs, as well as building connections between them in a single model, brings challenges to the design and training of the joint model. To address the problems, we propose a unified model for REG and REC, named UniRef. It unifies these two tasks with the carefully-designed Image-Region-Text Fusion layer (IRTF), which fuses the image, region and text via the image cross-attention and region cross-attention. Additionally, IRTF could generate pseudo input regions for the REC task to enable a uniform way for sharing the identical representation space across the REC and REG. We further propose Vision-conditioned Masked Language Modeling (VMLM) and Text-Conditioned Region Prediction (TRP) to pre-train UniRef model on multi-granular corpora. The VMLM and TRP are directly related to REG and REC, respectively, but could help each other. We conduct extensive experiments on three benchmark datasets, RefCOCO, RefCOCO+ and RefCOCOg. Experimental results show that our model outperforms previous state-of-the-art methods on both REG and REC.

While end-to-end neural machine translation (NMT) has achieved impressive progress, noisy input usually leads models to become fragile and unstable. Generating adversarial examples as the augmented data has been proved to be useful to alleviate this problem. Existing methods for adversarial example generation (AEG) are word-level or character-level, which ignore the ubiquitous phrase structure. In this paper, we propose a Phrase-level Adversarial Example Generation (PAEG) framework to enhance the robustness of the translation model. Our method further improves the gradient-based word-level AEG method by adopting a phrase-level substitution strategy. We verify our method on three benchmarks, including LDC Chinese-English, IWSLT14 German-English, and WMT14 English-German tasks. Experimental results demonstrate that our approach significantly improves translation performance and robustness to noise compared to previous strong baselines.