Machine Translation systems can produce different types of errors, some of which are characterized as critical or catastrophic due to the specific negative impact that they can have on users. In this paper we focus on one type of critical error: added toxicity. We evaluate and analyze added toxicity when translating a large evaluation dataset (HOLISTICBIAS, over 472k sentences, covering 13 demographic axes) from English into 164 languages. An automatic toxicity evaluation shows that added toxicity across languages varies from 0% to 5%. The output languages with the most added toxicity tend to be low-resource ones, and the demographic axes with the most added toxicity include sexual orientation, gender and sex, and ability. We also perform human evaluation on a subset of 8 translation directions, confirming the prevalence of true added toxicity. We use a measurement of the amount of source contribution to the translation, where a low source contribution implies hallucination, to interpret what causes toxicity. Making use of the input attributions allows us to explain toxicity, because the source contributions significantly correlate with toxicity for 84% of languages studied. Given our findings, our recommendations to reduce added toxicity are to curate training data to avoid mistranslations, mitigate hallucination and check unstable translations.
Identifying semantically equivalent sentences is important for many cross-lingual and mono-lingual NLP tasks. Current approaches to semantic equivalence take a loose, sentence-level approach to "equivalence," despite previous evidence that fine-grained differences and implicit content have an effect on human understanding (Roth and Anthonio, 2021) and system performance (Briakou and Carpuat, 2021). In this work, we introduce a novel, more sensitive method of characterizing semantic equivalence that leverages Abstract Meaning Representation graph structures. We develop an approach, which can be used with either gold or automatic AMR annotations, and demonstrate that our solution is in fact finer-grained than existing corpus filtering methods and more accurate at predicting strictly equivalent sentences than existing semantic similarity metrics. We suggest that our finer-grained measure of semantic equivalence could limit the workload in the task of human post-edited machine translation and in human evaluation of sentence similarity.
Language modelling and machine translation tasks mostly use subword or character inputs, but syllables are seldom used. Syllables provide shorter sequences than characters, require less-specialised extracting rules than morphemes, and their segmentation is not impacted by the corpus size. In this study, we first explore the potential of syllables for open-vocabulary language modelling in 21 languages. We use rule-based syllabification methods for six languages and address the rest with hyphenation, which works as a syllabification proxy. With a comparable perplexity, we show that syllables outperform characters and other subwords. Moreover, we study the importance of syllables on neural machine translation for a non-related and low-resource language-pair (Spanish--Shipibo-Konibo). In pairwise and multilingual systems, syllables outperform unsupervised subwords, and further morphological segmentation methods, when translating into a highly synthetic language with a transparent orthography (Shipibo-Konibo). Finally, we perform some human evaluation, and discuss limitations and opportunities.
Applying Reinforcement learning (RL) following maximum likelihood estimation (MLE) pre-training is a versatile method for enhancing neural machine translation (NMT) performance. However, recent work has argued that the gains produced by RL for NMT are mostly due to promoting tokens that have already received a fairly high probability in pre-training. We hypothesize that the large action space is a main obstacle to RL's effectiveness in MT, and conduct two sets of experiments that lend support to our hypothesis. First, we find that reducing the size of the vocabulary improves RL's effectiveness. Second, we find that effectively reducing the dimension of the action space without changing the vocabulary also yields notable improvement as evaluated by BLEU, semantic similarity, and human evaluation. Indeed, by initializing the network's final fully connected layer (that maps the network's internal dimension to the vocabulary dimension), with a layer that generalizes over similar actions, we obtain a substantial improvement in RL performance: 1.5 BLEU points on average.
Current learning models often struggle with human-like systematic generalization, particularly in learning compositional rules from limited data and extrapolating them to novel combinations. We introduce the Neural-Symbolic Recursive Machine (NSR), whose core is a Grounded Symbol System (GSS), allowing for the emergence of combinatorial syntax and semantics directly from training data. The NSR employs a modular design that integrates neural perception, syntactic parsing, and semantic reasoning. These components are synergistically trained through a novel deduction-abduction algorithm. Our findings demonstrate that NSR's design, imbued with the inductive biases of equivariance and compositionality, grants it the expressiveness to adeptly handle diverse sequence-to-sequence tasks and achieve unparalleled systematic generalization. We evaluate NSR's efficacy across four challenging benchmarks designed to probe systematic generalization capabilities: SCAN for semantic parsing, PCFG for string manipulation, HINT for arithmetic reasoning, and a compositional machine translation task. The results affirm NSR's superiority over contemporary neural and hybrid models in terms of generalization and transferability.
Arrival/Travel times for public transit exhibit variability on account of factors like seasonality, dwell times at bus stops, traffic signals, travel demand fluctuation etc. The developing world in particular is plagued by additional factors like lack of lane discipline, excess vehicles, diverse modes of transport and so on. This renders the bus arrival time prediction (BATP) to be a challenging problem especially in the developing world. A novel data-driven model based on recurrent neural networks (RNNs) is proposed for BATP (in real-time) in the current work. The model intelligently incorporates both spatial and temporal correlations in a unique (non-linear) fashion distinct from existing approaches. In particular, we propose a Gated Recurrent Unit (GRU) based Encoder-Decoder(ED) OR Seq2Seq RNN model (originally introduced for language translation) for BATP. The geometry of the dynamic real time BATP problem enables a nice fit with the Encoder-Decoder based RNN structure. We feed relevant additional synchronized inputs (from previous trips) at each step of the decoder (a feature classically unexplored in machine translation applications). Further motivated from accurately modelling congestion influences on travel time prediction, we additionally propose to use a bidirectional layer at the decoder (something unexplored in other time-series based ED application contexts). The effectiveness of the proposed algorithms is demonstrated on real field data collected from challenging traffic conditions. Our experiments indicate that the proposed method outperforms diverse existing state-of-art data-driven approaches proposed for the same problem.
We present FRMT, a new dataset and evaluation benchmark for Few-shot Region-aware Machine Translation, a type of style-targeted translation. The dataset consists of professional translations from English into two regional variants each of Portuguese and Mandarin Chinese. Source documents are selected to enable detailed analysis of phenomena of interest, including lexically distinct terms and distractor terms. We explore automatic evaluation metrics for FRMT and validate their correlation with expert human evaluation across both region-matched and mismatched rating scenarios. Finally, we present a number of baseline models for this task, and offer guidelines for how researchers can train, evaluate, and compare their own models. Our dataset and evaluation code are publicly available: https://bit.ly/frmt-task
Large pre-trained language models have brought remarkable progress in NLP. Pre-training and Fine-tuning have given state-of-art performance across tasks in text processing. Data Augmentation techniques have also helped build state-of-art models on low or zero resource tasks. Many works in the past have attempted at learning a single massively-multilingual machine translation model for zero-shot translation. Although those translation models are producing correct translations, the main challenge is those models are producing the wrong languages for zero-shot translation. This work and its results indicate that prompt conditioned large models do not suffer from off-target language errors i.e. errors arising due to translation to wrong languages. We empirically demonstrate the effectiveness of self-supervised pre-training and data augmentation for zero-shot multi-lingual machine translation.
The success of deep learning in natural language processing raises intriguing questions about the nature of linguistic meaning and ways in which it can be processed by natural and artificial systems. One such question has to do with subword segmentation algorithms widely employed in language modeling, machine translation, and other tasks since 2016. These algorithms often cut words into semantically opaque pieces, such as 'period', 'on', 't', and 'ist' in 'period|on|t|ist'. The system then represents the resulting segments in a dense vector space, which is expected to model grammatical relations among them. This representation may in turn be used to map 'period|on|t|ist' (English) to 'par|od|ont|iste' (French). Thus, instead of being modeled at the lexical level, translation is reformulated more generally as the task of learning the best bilingual mapping between the sequences of subword segments of two languages; and sometimes even between pure character sequences: 'p|e|r|i|o|d|o|n|t|i|s|t' $\rightarrow$ 'p|a|r|o|d|o|n|t|i|s|t|e'. Such subword segmentations and alignments are at work in highly efficient end-to-end machine translation systems, despite their allegedly opaque nature. The computational value of such processes is unquestionable. But do they have any linguistic or philosophical plausibility? I attempt to cast light on this question by reviewing the relevant details of the subword segmentation algorithms and by relating them to important philosophical and linguistic debates, in the spirit of making artificial intelligence more transparent and explainable.
With the recent advance in neural machine translation demonstrating its importance, research on quality estimation (QE) has been steadily progressing. QE aims to automatically predict the quality of machine translation (MT) output without reference sentences. Despite its high utility in the real world, there remain several limitations concerning manual QE data creation: inevitably incurred non-trivial costs due to the need for translation experts, and issues with data scaling and language expansion. To tackle these limitations, we present QUAK, a Korean-English synthetic QE dataset generated in a fully automatic manner. This consists of three sub-QUAK datasets QUAK-M, QUAK-P, and QUAK-H, produced through three strategies that are relatively free from language constraints. Since each strategy requires no human effort, which facilitates scalability, we scale our data up to 1.58M for QUAK-P, H and 6.58M for QUAK-M. As an experiment, we quantitatively analyze word-level QE results in various ways while performing statistical analysis. Moreover, we show that datasets scaled in an efficient way also contribute to performance improvements by observing meaningful performance gains in QUAK-M, P when adding data up to 1.58M.
Large multilingual models trained with self-supervision achieve state-of-the-art results in a wide range of natural language processing tasks. Self-supervised pretrained models are often fine-tuned on parallel data from one or multiple language pairs for machine translation. Multilingual fine-tuning improves performance on low-resource languages but requires modifying the entire model and can be prohibitively expensive. Training a new adapter on each language pair or training a single adapter on all language pairs without updating the pretrained model has been proposed as a parameter-efficient alternative. However, the former does not permit any sharing between languages, while the latter shares parameters for all languages and is susceptible to negative interference. In this paper, we propose training language-family adapters on top of mBART-50 to facilitate cross-lingual transfer. Our approach outperforms related baselines, yielding higher translation scores on average when translating from English to 17 different low-resource languages. We also show that language-family adapters provide an effective method to translate to languages unseen during pretraining.
Recent advances on the Vector Space Model have significantly improved some NLP applications such as neural machine translation and natural language generation. Although word co-occurrences in context have been widely used in counting-/predicting-based distributional models, the role of syntactic dependencies in deriving distributional semantics has not yet been thoroughly investigated. By comparing various Vector Space Models in detecting synonyms in TOEFL, we systematically study the salience of syntactic dependencies in accounting for distributional similarity. We separate syntactic dependencies into different groups according to their various grammatical roles and then use context-counting to construct their corresponding raw and SVD-compressed matrices. Moreover, using the same training hyperparameters and corpora, we study typical neural embeddings in the evaluation. We further study the effectiveness of injecting human-compiled semantic knowledge into neural embeddings on computing distributional similarity. Our results show that the syntactically conditioned contexts can interpret lexical semantics better than the unconditioned ones, whereas retrofitting neural embeddings with semantic knowledge can significantly improve synonym detection.
Buddhism is an influential religion with a long-standing history and profound philosophy. Nowadays, more and more people worldwide aspire to learn the essence of Buddhism, attaching importance to Buddhism dissemination. However, Buddhist scriptures written in classical Chinese are obscure to most people and machine translation applications. For instance, general Chinese-English neural machine translation (NMT) fails in this domain. In this paper, we proposed a novel approach to building a practical NMT model for Buddhist scriptures. The performance of our translation pipeline acquired highly promising results in ablation experiments under three criteria.
Large language models such as GPT-3 (Brown et al., 2020) can perform arbitrary tasks without undergoing fine-tuning after being prompted with only a few labeled examples. An arbitrary task can be reformulated as a natural language prompt, and a language model can be asked to generate the completion, indirectly performing the task in a paradigm known as prompt-based learning. To date, emergent prompt-based learning capabilities have mainly been demonstrated for unidirectional language models. However, bidirectional language models pre-trained on denoising objectives such as masked language modeling produce stronger learned representations for transfer learning. This motivates the possibility of prompting bidirectional models, but their pre-training objectives have made them largely incompatible with the existing prompting paradigm. We present SAP (Sequential Autoregressive Prompting), a technique that enables the prompting of bidirectional models. Utilizing the machine translation task as a case study, we prompt the bidirectional mT5 model (Xue et al., 2021) with SAP and demonstrate its few-shot and zero-shot translations outperform the few-shot translations of unidirectional models like GPT-3 and XGLM (Lin et al., 2021), despite mT5's approximately 50% fewer parameters. We further show SAP is effective on question answering and summarization. For the first time, our results demonstrate prompt-based learning is an emergent property of a broader class of language models, rather than only unidirectional models.
This report presents an automatic evaluation of the general machine translation task of the Seventh Conference on Machine Translation (WMT22). It evaluates a total of 185 systems for 21 translation directions including high-resource to low-resource language pairs and from closely related to distant languages. This large-scale automatic evaluation highlights some of the current limits of state-of-the-art machine translation systems. It also shows how automatic metrics, namely chrF, BLEU, and COMET, can complement themselves to mitigate their own limits in terms of interpretability and accuracy.
In this paper, we examine the development and usage of six low-resource machine translation systems translating between the Ukrainian language and each of the official languages of the Baltic states. We developed these systems in reaction to the escalating Ukrainian refugee crisis caused by the Russian military aggression in Ukraine in the hope that they might be helpful for refugees and public administrations. Now, two months after MT systems were made public, we analyze their usage patterns and statistics. Our findings show that the Latvian-Ukrainian and Lithuanian-Ukrainian systems are integrated into the public services of Baltic states, leading to more than 127 million translated sentences for the Lithuanian-Ukrainian system. Motivated by these findings, we further enhance our MT systems by better Ukrainian toponym translation and publish an improved version of the Lithuanian-Ukrainian system.
One of the vital breakthroughs in the history of machine translation is the development of the Transformer model. Not only it is revolutionary for various translation tasks, but also for a majority of other NLP tasks. In this paper, we aim at a Transformer-based system that is able to translate a source sentence in German to its counterpart target sentence in English. We perform the experiments on the news commentary German-English parallel sentences from the WMT'13 dataset. In addition, we investigate the effect of the inclusion of additional general-domain data in training from the IWSLT'16 dataset to improve the Transformer model performance. We find that including the IWSLT'16 dataset in training helps achieve a gain of 2 BLEU score points on the test set of the WMT'13 dataset. Qualitative analysis is introduced to analyze how the usage of general-domain data helps improve the quality of the produced translation sentences.
Despite the fact that multilingual agreement (MA) has shown its importance for multilingual neural machine translation (MNMT), current methodologies in the field have two shortages: (i) require parallel data between multiple language pairs, which is not always realistic and (ii) optimize the agreement in an ambiguous direction, which hampers the translation performance. We present \textbf{B}idirectional \textbf{M}ultilingual \textbf{A}greement via \textbf{S}witched \textbf{B}ack-\textbf{t}ranslation (\textbf{BMA-SBT}), a novel and universal multilingual agreement framework for fine-tuning pre-trained MNMT models, which (i) exempts the need for aforementioned parallel data by using a novel method called switched BT that creates synthetic text written in another source language using the translation target and (ii) optimizes the agreement bidirectionally with the Kullback-Leibler Divergence loss. Experiments indicate that BMA-SBT clearly improves the strong baselines on the task of MNMT with three benchmarks: TED Talks, News, and Europarl. In-depth analyzes indicate that BMA-SBT brings additive improvements to the conventional BT method.
We hypothesize that existing sentence-level machine translation (MT) metrics become less effective when the human reference contains ambiguities. To verify this hypothesis, we present a very simple method for extending pretrained metrics to incorporate context at the document level. We apply our method to three popular metrics, BERTScore, Prism, and COMET, and to the reference free metric COMET-QE. We evaluate the extended metrics on the WMT 2021 metrics shared task using the provided MQM annotations. Our results show that the extended metrics outperform their sentence-level counterparts in about 85% of the tested conditions, when excluding results on low-quality human references. Additionally, we show that our document-level extension of COMET-QE dramatically improves its accuracy on discourse phenomena tasks, outperforming a dedicated baseline by up to 6.1%. Our experimental results support our initial hypothesis and show that a simple extension of the metrics permits them to take advantage of context to resolve ambiguities in the reference.
In intelligent manufacturing, the quality of machine translation engineering drawings will directly affect its manufacturing accuracy. Currently, most of the work is manually translated, greatly reducing production efficiency. This paper proposes an automatic translation method for welded structural engineering drawings based on Cyclic Generative Adversarial Networks (CycleGAN). The CycleGAN network model of unpaired transfer learning is used to learn the feature mapping of real welding engineering drawings to realize automatic translation of engineering drawings. U-Net and PatchGAN are the main network for the generator and discriminator, respectively. Based on removing the identity mapping function, a high-dimensional sparse network is proposed to replace the traditional dense network for the Cyclegan generator to improve noise robustness. Increase the residual block hidden layer to increase the resolution of the generated graph. The improved and fine-tuned network models are experimentally validated, computing the gap between real and generated data. It meets the welding engineering precision standard and solves the main problem of low drawing recognition efficiency in the welding manufacturing process. The results show. After training with our model, the PSNR, SSIM and MSE of welding engineering drawings reach about 44.89%, 99.58% and 2.11, respectively, which are superior to traditional networks in both training speed and accuracy.