The COMET metric has blazed a trail in the machine translation community, given its strong correlation with human judgements of translation quality. Its success stems from being a modified pre-trained multilingual model finetuned for quality assessment. However, it being a machine learning model also gives rise to a new set of pitfalls that may not be widely known. We investigate these unexpected behaviours from three aspects: 1) technical: obsolete software versions and compute precision; 2) data: empty content, language mismatch, and translationese at test time as well as distribution and domain biases in training; 3) usage and reporting: multi-reference support and model referencing in the literature. All of these problems imply that COMET scores are not comparable between papers or even technical setups and we put forward our perspective on fixing each issue. Furthermore, we release the sacreCOMET package that can generate a signature for the software and model configuration as well as an appropriate citation. The goal of this work is to help the community make more sound use of the COMET metric.
Machine translation models are still inappropriate for translating chats, despite the popularity of translation software and plug-in applications. The complexity of dialogues poses significant challenges and can hinder crosslingual communication. Instead of pursuing a flawless translation system, a more practical approach would be to issue warning messages about potential mistranslations to reduce confusion. However, it is still unclear how individuals perceive these warning messages and whether they benefit the crowd. This paper tackles to investigate this question and demonstrates the warning messages' contribution to making chat translation systems effective.
With the development of artificial intelligence, particularly the success of Large Language Models (LLMs), the quantity and quality of automatically generated stories have significantly increased. This has led to the need for automatic story evaluation to assess the generative capabilities of computing systems and analyze the quality of both automatic-generated and human-written stories. Evaluating a story can be more challenging than other generation evaluation tasks. While tasks like machine translation primarily focus on assessing the aspects of fluency and accuracy, story evaluation demands complex additional measures such as overall coherence, character development, interestingness, etc. This requires a thorough review of relevant research. In this survey, we first summarize existing storytelling tasks, including text-to-text, visual-to-text, and text-to-visual. We highlight their evaluation challenges, identify various human criteria to measure stories, and present existing benchmark datasets. Then, we propose a taxonomy to organize evaluation metrics that have been developed or can be adopted for story evaluation. We also provide descriptions of these metrics, along with the discussion of their merits and limitations. Later, we discuss the human-AI collaboration for story evaluation and generation. Finally, we suggest potential future research directions, extending from story evaluation to general evaluations.
With the growing Deaf and Hard of Hearing population worldwide and the persistent shortage of certified sign language interpreters, there is a pressing need for an efficient, signs-driven, integrated end-to-end translation system, from sign to gloss to text and vice-versa. There has been a wealth of research on machine translations and related reviews. However, there are few works on sign language machine translation considering the particularity of the language being continuous and dynamic. This paper aims to address this void, providing a retrospective analysis of the temporal evolution of sign language machine translation algorithms and a taxonomy of the Transformers architectures, the most used approach in language translation. We also present the requirements of a real-time Quality-of-Service sign language ma-chine translation system underpinned by accurate deep learning algorithms. We propose future research directions for sign language translation systems.
In this provocation, we discuss the English dominance of the AI research community, arguing that the requirement for English language publishing upholds and reinforces broader regimes of extraction in AI. While large language models and machine translation have been celebrated as a way to break down barriers, we regard their use as a symptom of linguistic exclusion of scientists and potential readers. We propose alternative futures for a healthier publishing culture, organized around three themes: administering conferences in the languages of the country in which they are held, instructing peer reviewers not to adjudicate the language appropriateness of papers, and offering opportunities to publish and present in multiple languages. We welcome new translations of this piece. Please contact the authors if you would like to contribute one.
Autoregressive Sequence-To-Sequence models are the foundation of many Deep Learning achievements in major research fields such as Vision and Natural Language Processing. Despite that, they still present significant limitations. For instance, when errors occur in the early steps of the prediction, the whole output is severely affected. Such reliance on previously predicted tokens and the inherent computational unfriendliness of sequential algorithms, motivated researchers to explore different architectures and methods in the search for bidirectional approaches. In this work, we introduce the Bidirectional Awareness Induction (BAI), a training method that leverages a subset of elements in the network, the Pivots, to perform bidirectional learning without breaking the autoregressive constraints. To showcase its flexibility, we apply the method to three architectures, the Transformer, ExpansionNet v2 and GPT, then perform experiments over three tasks. Experimental results showcase BAI's effectiveness on all selected tasks and architectures. In particular, we observed an increase of up to 2.4 CIDEr in Image-Captioning, 4.96 BLEU in Neural Machine Translation, and 1.16 ROUGE in Text Summarization compared to the respective baselines. Notably, BAI not only has a positive impact on models trained from scratch but on pre-trained models as well. Such an aspect, combined with the absence of architectural requirements synergizes well with the current trend of LLMs.
Machine Translation of Culture-Specific Items (CSIs) poses significant challenges. Recent work on CSI translation has shown some success using Large Language Models (LLMs) to adapt to different languages and cultures; however, a deeper analysis is needed to examine the benefits and pitfalls of each method. In this paper, we introduce the ChineseMenuCSI dataset, the largest for Chinese-English menu corpora, annotated with CSI vs Non-CSI labels and a fine-grained test set. We define three levels of CSI figurativeness for a more nuanced analysis and develop a novel methodology for automatic CSI identification, which outperforms GPT-based prompts in most categories. Importantly, we are the first to integrate human translation theories into LLM-driven translation processes, significantly improving translation accuracy, with COMET scores increasing by up to 7 points.
The main task of the KGQA system (Knowledge Graph Question Answering) is to convert user input questions into query syntax (such as SPARQL). With the rise of modern popular encoders and decoders like Transformer and ConvS2S, many scholars have shifted the research direction of SPARQL generation to the Neural Machine Translation (NMT) architecture or the generative AI field of Text-to-SPARQL. In NMT-based QA systems, the system treats knowledge base query syntax as a language. It uses NMT-based translation models to translate natural language questions into query syntax. Scholars use popular architectures equipped with cross-attention, such as Transformer, ConvS2S, and BiLSTM, to train translation models for query syntax. To achieve better query results, this paper improved the ConvS2S encoder and added multi-head attention from the Transformer, proposing a Multi-Head Conv encoder (MHC encoder) based on the n-gram language model. The principle is to use convolutional layers to capture local hidden features in the input sequence with different receptive fields, using multi-head attention to calculate dependencies between them. Ultimately, we found that the translation model based on the Multi-Head Conv encoder achieved better performance than other encoders, obtaining 76.52\% and 83.37\% BLEU-1 (BiLingual Evaluation Understudy) on the QALD-9 and LC-QuAD-1.0 datasets, respectively. Additionally, in the end-to-end system experiments on the QALD-9 and LC-QuAD-1.0 datasets, we achieved leading results over other KGQA systems, with Macro F1-measures reaching 52\% and 66\%, respectively. Moreover, the experimental results show that with limited computational resources, if one possesses an excellent encoder-decoder architecture and cross-attention, experts and scholars can achieve outstanding performance equivalent to large pre-trained models using only general embeddings.
Sign language translation has historically been peripheral to mainstream machine translation research. In order to help converge the fields, we introduce FLEURS-ASL, an extension of the multiway parallel benchmarks FLORES (for text) and FLEURS (for speech) to support their first sign language (as video), American Sign Language, translated by 5 Certified Deaf Interpreters. FLEURS-ASL can be used to evaluate a variety of tasks -- primarily sentence- and discourse-level translation -- between ASL and 200 other languages as text, or 102 languages as speech. We provide baselines for tasks from ASL to English text using a unified modeling approach that incorporates timestamp tokens and previous text tokens in a 34-second context window, trained on random video clips from YouTube-ASL. This model meets or exceeds the performance of phrase-level baselines while supporting a multitude of new tasks. We also use FLEURS-ASL to show that multimodal frontier models have virtually no understanding of ASL, underscoring the importance of including sign languages in standard evaluation suites.
This research explores the development of multimodal vision-language models for image retrieval in low-resource languages, specifically Azerbaijani. Existing vision-language models primarily support high-resource languages, and fine-tuning them remains computationally demanding. To address challenges in vision-language retrieval for low-resource languages, we integrated the CLIP model architecture and employed several techniques to balance computational efficiency with performance. These techniques include synthetic data generation through machine translation, image augmentation, and further training the attention mechanisms of transformer-based models with domain-specific data. We integrated Multilingual BERT as a text encoder with image encoders like ResNet50, EfficientNet0, Vision Transformer (ViT), and Tiny Swin Transformer. Our study found that models like EfficientNet0 and Tiny Swin Transformer perform best on the datasets they were trained on, such as COCO, Flickr30k, and Flickr8k. Augmentation techniques boosted EfficientNet0 MAP on Flickr30k from 0.84 to 0.87 and ResNet50 MAP on MSCOCO from 0.70 to 0.80, contributing to a new state of the art in vision-language retrieval. We share our configurations and results to support further research. Code and pre-trained models are available at https://github.com/aliasgerovs/azclip.
Annually, at the Conference of Machine Translation (WMT), the Metrics Shared Task organizers conduct the meta-evaluation of Machine Translation (MT) metrics, ranking them according to their correlation with human judgments. Their results guide researchers toward enhancing the next generation of metrics and MT systems. With the recent introduction of neural metrics, the field has witnessed notable advancements. Nevertheless, the inherent opacity of these metrics has posed substantial challenges to the meta-evaluation process. This work highlights two issues with the meta-evaluation framework currently employed in WMT, and assesses their impact on the metrics rankings. To do this, we introduce the concept of sentinel metrics, which are designed explicitly to scrutinize the meta-evaluation process's accuracy, robustness, and fairness. By employing sentinel metrics, we aim to validate our findings, and shed light on and monitor the potential biases or inconsistencies in the rankings. We discover that the present meta-evaluation framework favors two categories of metrics: i) those explicitly trained to mimic human quality assessments, and ii) continuous metrics. Finally, we raise concerns regarding the evaluation capabilities of state-of-the-art metrics, emphasizing that they might be basing their assessments on spurious correlations found in their training data.
Despite the recent popularity of Large Language Models (LLMs) in Machine Translation (MT), their performance in low-resource languages (LRLs) still lags significantly behind Neural Machine Translation (NMT) models. In this work, we explore what it would take to adapt LLMs for the low-resource setting. Particularly, we re-examine the role of two factors: a) the importance and application of parallel data, and b) diversity in Supervised Fine-Tuning (SFT). Recently, parallel data has seen reduced use in adapting LLMs for MT, while data diversity has been embraced to promote transfer across languages and tasks. However, for low-resource LLM-MT, we show that the opposite is true for both considerations: a) parallel data is critical during both pre-training and SFT; b) diversity tends to cause interference instead of transfer. Our experiments with three LLMs across two low-resourced language groups -- Indigenous American and North-East Indian -- reveal consistent trends, underscoring the generalizability of our findings. We believe these insights will be valuable for scaling to massively multilingual LLM-MT models that can effectively serve LRLs.
The deep learning language of choice these days is Python; measured by factors such as available libraries and technical support, it is hard to beat. At the same time, software written in lower-level programming languages like C++ retain advantages in speed. We describe a Python interface to Marian NMT, a C++-based training and inference toolkit for sequence-to-sequence models, focusing on machine translation. This interface enables models trained with Marian to be connected to the rich, wide range of tools available in Python. A highlight of the interface is the ability to compute state-of-the-art COMET metrics from Python but using Marian's inference engine, with a speedup factor of up to 7.8$\times$ the existing implementations. We also briefly spotlight a number of other integrations, including Jupyter notebooks, connection with prebuilt models, and a web app interface provided with the package. PyMarian is available in PyPI via $\texttt{pip install pymarian}$.
Recent advancements in neural machine translation (NMT) have revolutionized the field, yet the dependency on extensive parallel corpora limits progress for low-resource languages and domains. Cross-lingual transfer learning offers a promising solution by utilizing data from high-resource languages but often struggles with in-domain NMT. This paper investigates zero-shot cross-lingual domain adaptation for NMT, focusing on the impact of domain specification and linguistic factors on transfer effectiveness. Using English as the source language and Spanish for fine-tuning, we evaluate multiple target languages, including Portuguese, Italian, French, Czech, Polish, and Greek. We demonstrate that both language-specific and domain-specific factors influence transfer effectiveness, with domain characteristics playing a crucial role in determining cross-domain transfer potential. We also explore the feasibility of zero-shot cross-lingual cross-domain transfer, providing insights into which domains are more responsive to transfer and why. Our results show the importance of well-defined domain boundaries and transparency in experimental setups for in-domain transfer learning.
Back translation, as a technique for extending a dataset, is widely used by researchers in low-resource language translation tasks. It typically translates from the target to the source language to ensure high-quality translation results. This paper proposes a novel way of utilizing a monolingual corpus on the source side to assist Neural Machine Translation (NMT) in low-resource settings. We realize this concept by employing a Generative Adversarial Network (GAN), which augments the training data for the discriminator while mitigating the interference of low-quality synthetic monolingual translations with the generator. Additionally, this paper integrates Translation Memory (TM) with NMT, increasing the amount of data available to the generator. Moreover, we propose a novel procedure to filter the synthetic sentence pairs during the augmentation process, ensuring the high quality of the data.
Neural Machine Translation (NMT) models have traditionally used Sinusoidal Positional Embeddings (PEs), which often struggle to capture long-range dependencies and are less efficient for handling extended context or document-level translation tasks. This work addresses the challenge of transitioning pre-trained NMT models from absolute sinusoidal PEs to relative PEs, such as Rotary Positional Embeddings (ROPE) and Attention with Linear Biases (ALIBI), without compromising performance. We demonstrate that parameter-efficient fine-tuning, using only a small amount of high-quality data, can successfully facilitate this transition. Experimental results indicate that switching from sinusoidal to relative PEs results in competitive translation quality on sentence-level evaluation benchmarks. Additionally, models trained with ROPE consistently outperform those using ALIBI and Sinusoidal PEs on document-level benchmarks across both string-based metrics and qualitative evaluations. Moreover, we find that a small amount of long-context data in a few languages is sufficient for cross-lingual length generalization, thereby inducing long-context capabilities.
This paper introduces two multilingual systems, IKUN and IKUN-C, developed for the general machine translation task in WMT24. IKUN and IKUN-C represent an open system and a constrained system, respectively, built on Llama-3-8b and Mistral-7B-v0.3. Both systems are designed to handle all 11 language directions using a single model. According to automatic evaluation metrics, IKUN-C achieved 6 first-place and 3 second-place finishes among all constrained systems, while IKUN secured 1 first-place and 2 second-place finishes across both open and constrained systems. These encouraging results suggest that large language models (LLMs) are nearing the level of proficiency required for effective multilingual machine translation. The systems are based on a two-stage approach: first, continuous pre-training on monolingual data in 10 languages, followed by fine-tuning on high-quality parallel data for 11 language directions. The primary difference between IKUN and IKUN-C lies in their monolingual pre-training strategy. IKUN-C is pre-trained using constrained monolingual data, whereas IKUN leverages monolingual data from the OSCAR dataset. In the second phase, both systems are fine-tuned on parallel data sourced from NTREX, Flores, and WMT16-23 for all 11 language pairs.
As part of the Open Language Data Initiative shared tasks, we have expanded the FLORES+ evaluation set to include Emakhuwa, a low-resource language widely spoken in Mozambique. We translated the dev and devtest sets from Portuguese into Emakhuwa, and we detail the translation process and quality assurance measures used. Our methodology involved various quality checks, including post-editing and adequacy assessments. The resulting datasets consist of multiple reference sentences for each source. We present baseline results from training a Neural Machine Translation system and fine-tuning existing multilingual translation models. Our findings suggest that spelling inconsistencies remain a challenge in Emakhuwa. Additionally, the baseline models underperformed on this evaluation set, underscoring the necessity for further research to enhance machine translation quality for Emakhuwa. The data is publicly available at https://huggingface.co/datasets/LIACC/Emakhuwa-FLORES.
Subword regularization, used widely in NLP, improves model performance by reducing the dependency on exact tokenizations, augmenting the training corpus, and exposing the model to more unique contexts during training. BPE and MaxMatch, two popular subword tokenization schemes, have stochastic dropout regularization variants. However, there has not been an analysis of the distributions formed by them. We show that these stochastic variants are heavily biased towards a small set of tokenizations per word. If the benefits of subword regularization are as mentioned, we hypothesize that biasedness artificially limits the effectiveness of these schemes. Thus, we propose an algorithm to uniformly sample tokenizations that we use as a drop-in replacement for the stochastic aspects of existing tokenizers, and find that it improves machine translation quality.
Recent advancements in NLP have resulted in models with specialized strengths, such as processing multimodal inputs or excelling in specific domains. However, real-world tasks, like multimodal translation, often require a combination of these strengths, such as handling both translation and image processing. While individual translation and vision models are powerful, they typically lack the ability to perform both tasks in a single system. Combining these models poses challenges, particularly due to differences in their vocabularies, which limit the effectiveness of traditional ensemble methods to post-generation techniques like N-best list re-ranking. In this work, we propose a novel zero-shot ensembling strategy that allows for the integration of different models during the decoding phase without the need for additional training. Our approach re-ranks beams during decoding by combining scores at the word level, using heuristics to predict when a word is completed. We demonstrate the effectiveness of this method in machine translation scenarios, showing that it enables the generation of translations that are both speech- and image-aware while also improving overall translation quality (We will release the code upon paper acceptance.).