We present our systems participated in the VLSP 2022 machine translation shared task. In the shared task this year, we participated in both translation tasks, i.e., Chinese-Vietnamese and Vietnamese-Chinese translations. We build our systems based on the neural-based Transformer model with the powerful multilingual denoising pre-trained model mBART. The systems are enhanced by a sampling method for backtranslation, which leverage large scale available monolingual data. Additionally, several other methods are applied to improve the translation quality including ensembling and postprocessing. We achieve 38.9 BLEU on ChineseVietnamese and 38.0 BLEU on VietnameseChinese on the public test sets, which outperform several strong baselines.

Many machine translation toolkits make use of a data preparation step wherein raw data is transformed into a tensor format that can be used directly by the trainer. This preparation step is increasingly at odds with modern research and development practices because this process produces a static, unchangeable version of the training data, making common training-time needs difficult (e.g., subword sampling), time-consuming (preprocessing with large data can take days), expensive (e.g., disk space), and cumbersome (managing experiment combinatorics). We propose an alternative approach that separates the generation of data from the consumption of that data. In this approach, there is no separate pre-processing step; data generation produces an infinite stream of permutations of the raw training data, which the trainer tensorizes and batches as it is consumed. Additionally, this data stream can be manipulated by a set of user-definable operators that provide on-the-fly modifications, such as data normalization, augmentation or filtering. We release an open-source toolkit, SOTASTREAM, that implements this approach: https://github.com/marian-nmt/sotastream. We show that it cuts training time, adds flexibility, reduces experiment management complexity, and reduces disk space, all without affecting the accuracy of the trained models.

Automatic evaluation of machine translation (MT) is a critical tool driving the rapid iterative development of MT systems. While considerable progress has been made on estimating a single scalar quality score, current metrics lack the informativeness of more detailed schemes that annotate individual errors, such as Multidimensional Quality Metrics (MQM). In this paper, we help fill this gap by proposing AutoMQM, a prompting technique which leverages the reasoning and in-context learning capabilities of large language models (LLMs) and asks them to identify and categorize errors in translations. We start by evaluating recent LLMs, such as PaLM and PaLM-2, through simple score prediction prompting, and we study the impact of labeled data through in-context learning and finetuning. We then evaluate AutoMQM with PaLM-2 models, and we find that it improves performance compared to just prompting for scores (with particularly large gains for larger models) while providing interpretability through error spans that align with human annotations.

A key challenge in professional fact-checking is its limited scalability in relation to the magnitude of false information. While many Natural Language Processing (NLP) tools have been proposed to enhance fact-checking efficiency and scalability, both academic research and fact-checking organizations report limited adoption of such tooling due to insufficient alignment with fact-checker practices, values, and needs. To address this gap, we investigate a co-design method, Matchmaking for AI, which facilitates fact-checkers, designers, and NLP researchers to collaboratively discover what fact-checker needs should be addressed by technology and how. Our co-design sessions with 22 professional fact-checkers yielded a set of 11 novel design ideas. They assist in information searching, processing, and writing tasks for efficient and personalized fact-checking; help fact-checkers proactively prepare for future misinformation; monitor their potential biases; and support internal organization collaboration. Our work offers implications for human-centered fact-checking research and practice and AI co-design research.

A large catalogue size is one of the central challenges in training recommendation models: a large number of items makes them memory and computationally inefficient to compute scores for all items during training, forcing these models to deploy negative sampling. However, negative sampling increases the proportion of positive interactions in the training data, and therefore models trained with negative sampling tend to overestimate the probabilities of positive interactions a phenomenon we call overconfidence. While the absolute values of the predicted scores or probabilities are not important for the ranking of retrieved recommendations, overconfident models may fail to estimate nuanced differences in the top-ranked items, resulting in degraded performance. In this paper, we show that overconfidence explains why the popular SASRec model underperforms when compared to BERT4Rec. This is contrary to the BERT4Rec authors explanation that the difference in performance is due to the bi-directional attention mechanism. To mitigate overconfidence, we propose a novel Generalised Binary Cross-Entropy Loss function (gBCE) and theoretically prove that it can mitigate overconfidence. We further propose the gSASRec model, an improvement over SASRec that deploys an increased number of negatives and the gBCE loss. We show through detailed experiments on three datasets that gSASRec does not exhibit the overconfidence problem. As a result, gSASRec can outperform BERT4Rec (e.g. +9.47% NDCG on the MovieLens-1M dataset), while requiring less training time (e.g. -73% training time on MovieLens-1M). Moreover, in contrast to BERT4Rec, gSASRec is suitable for large datasets that contain more than 1 million items.

This work presents our efforts to reproduce the results of the human evaluation experiment presented in the paper of Vamvas and Sennrich (2022), which evaluated an automatic system detecting over- and undertranslations (translations containing more or less information than the original) in machine translation (MT) outputs. Despite the high quality of the documentation and code provided by the authors, we discuss some problems we found in reproducing the exact experimental setup and offer recommendations for improving reproducibility. Our replicated results generally confirm the conclusions of the original study, but in some cases, statistically significant differences were observed, suggesting a high variability of human annotation.

Translation Quality Estimation (QE) is the task of predicting the quality of machine translation (MT) output without any reference. This task has gained increasing attention as an important component in the practical applications of MT. In this paper, we first propose XLMRScore, which is a cross-lingual counterpart of BERTScore computed via the XLM-RoBERTa (XLMR) model. This metric can be used as a simple unsupervised QE method, while employing it results in two issues: firstly, the untranslated tokens leading to unexpectedly high translation scores, and secondly, the issue of mismatching errors between source and hypothesis tokens when applying the greedy matching in XLMRScore. To mitigate these issues, we suggest replacing untranslated words with the unknown token and the cross-lingual alignment of the pre-trained model to represent aligned words closer to each other, respectively. We evaluate the proposed method on four low-resource language pairs of WMT21 QE shared task, as well as a new English-Farsi test dataset introduced in this paper. Experiments show that our method could get comparable results with the supervised baseline for two zero-shot scenarios, i.e., with less than 0.01 difference in Pearson correlation, while outperforming unsupervised rivals in all the low-resource language pairs for above 8%, on average.

Recent studies have shown that the multi-encoder models are agnostic to the choice of context, and the context encoder generates noise which helps improve the models in terms of BLEU score. In this paper, we further explore this idea by evaluating with context-aware pronoun translation test set by training multi-encoder models trained on three different context settings viz, previous two sentences, random two sentences, and a mix of both as context. Specifically, we evaluate the models on the ContraPro test set to study how different contexts affect pronoun translation accuracy. The results show that the model can perform well on the ContraPro test set even when the context is random. We also analyze the source representations to study whether the context encoder generates noise. Our analysis shows that the context encoder provides sufficient information to learn discourse-level information. Additionally, we observe that mixing the selected context (the previous two sentences in this case) and the random context is generally better than the other settings.

In machine translation tasks, the relationship between model complexity and performance is often presumed to be linear, driving an increase in the number of parameters and consequent demands for computational resources like multiple GPUs. To explore this assumption, this study systematically investigates the effects of hyperparameters through ablation on a sequence-to-sequence machine translation pipeline, utilizing a single NVIDIA A100 GPU. Contrary to expectations, our experiments reveal that combinations with the most parameters were not necessarily the most effective. This unexpected insight prompted a careful reduction in parameter sizes, uncovering "sweet spots" that enable training sophisticated models on a single GPU without compromising translation quality. The findings demonstrate an intricate relationship between hyperparameter selection, model size, and computational resource needs. The insights from this study contribute to the ongoing efforts to make machine translation more accessible and cost-effective, emphasizing the importance of precise hyperparameter tuning over mere scaling.

Language models that are sensitive to external context can more effectively capture the speaking patterns of individuals with specific characteristics or in particular environments. However, obtaining and leveraging such annotations can be challenging. In this work, we show how to leverage rich character and film annotations to personalise language models in a scalable manner. Our best model can reduce perplexity by up to 6.5% compared to a parameter-matched language model. Our approach performs on par with speaker-specific fine-tuning when the fine-tuning data (i.e. past dialogue) for individual speakers is available. On top of that, it also generalises well to a scenario with no such data, relying on combinations of demographic characteristics expressed via metadata. Our findings are consistent across two corpora, one of which is also a contribution of this paper: Cornell-rich contains rich manual annotations for 863 speaking characters from the Cornell Movie Dialog Corpus, including features such as characteristic quotes and character descriptions, along with six automatically extracted metadata features for over 95% of the featured films. Finally, we also present a cost-benefit analysis highlighting which annotations are most cost-effective in reducing perplexity.