Search

The encoder-decoder with attention model has become the state of the art for machine translation. However, more investigations are still needed to understand the internal mechanism of this end-to-end model. In this paper, we focus on how neural machine translation (NMT) models consider source information while decoding. We propose a numerical measurement of source context dependency in the NMT models and analyze the behaviors of the NMT decoder with this measurement under several circumstances. Experimental results show that this measurement is an appropriate estimate for source context dependency and consistent over different domains.

A prerequisite for training corpus-based machine translation (MT) systems – either Statistical MT (SMT) or Neural MT (NMT) – is the availability of high-quality parallel data. This is arguably more important today than ever before, as NMT has been shown in many studies to outperform SMT, but mostly when large parallel corpora are available; in cases where data is limited, SMT can still outperform NMT. Recently researchers have shown that back-translating monolingual data can be used to create synthetic parallel corpora, which in turn can be used in combination with authentic parallel data to train a high-quality NMT system. Given that large collections of new parallel text become available only quite rarely, back-translation has become the norm when building state-of-the-art NMT systems, especially in resource-poor scenarios. However, we assert that there are many unknown factors regarding the actual effects of back-translated data on the translation capabilities of an NMT model. Accordingly, in this work we investigate how using back-translated data as a training corpus – both as a separate standalone dataset as well as combined with human-generated parallel data – affects the performance of an NMT model. We use incrementally larger amounts of back-translated data to train a range of NMT systems for German-to-English, and analyse the resulting translation performance.

We describe the European Language Resources Infrastructure project, whose main aim is the provision of an infrastructure to help collect, prepare and share language resources that can in turn improve translation services in Europe.

This short presentation introduces ModernMT: an open-source project that integrates real-time adaptive neural machine translation into a single easy-to-use product.

MultiMT is an European Research Council Starting Grant whose aim is to devise data, methods and algorithms to exploit multi-modal information (images, audio, metadata) for context modelling in machine translation and other cross-lingual tasks. The project draws upon different research fields including natural language processing, computer vision, speech processing and machine learning.

Telugu is the fifteenth most commonly spoken language in the world with an estimated reach of 75 million people in the Indian subcontinent. At the same time, it is a severely low resourced language. In this paper, we present work on English–Telugu general domain machine translation (MT) systems using small amounts of parallel data. The baseline statistical (SMT) and neural MT (NMT) systems do not yield acceptable translation quality, mostly due to limited resources. However, the use of synthetic parallel data (generated using back translation, based on an NMT engine) significantly improves translation quality and allows NMT to outperform SMT. We extend back translation and propose a new, iterative data augmentation (IDA) method. Filtering of synthetic data and IDA both further boost translation quality of our final NMT systems, as measured by BLEU scores on all test sets and based on state-of-the-art human evaluation.

With the expansion of MT usage at TransPerfect, we have developed an implementation strategy that involves continuous work with linguists on a wide variety of MT-related tasks. Today, MT undeniably plays a big role in translators’ lives. As an internal linguist at TransPerfect, I have experienced it in my everyday work. A big part of it is now related to MT and these tasks include not only MT post-editing, but also MT evaluation and improvement.

We will present our solution to replace the usage of publicly available machine translation (MT) services in companies where privacy and confidentiality are key. Our MT portal can translate across a variety of languages using neural machine translation, and supports an extensive number of file types. Corporations are using it to enable multilingual communication everywhere.

The objective of this paper is to analyse some aspects related to the practice of post-editing services in the current translation market in Spain. To this aim, some quantitative data collected through an online survey and concerning the experience and opinion of professional translators regarding post-editing will be shown.

We present an approach to neural machine translation (NMT) that supports multiple domains in a single model and allows switching between the domains when translating. The core idea is to treat text domains as distinct languages and use multilingual NMT methods to create multi-domain translation systems; we show that this approach results in significant translation quality gains over fine-tuning. We also explore whether the knowledge of pre-specified text domains is necessary; turns out that it is after all, but also that when it is not known quite high translation quality can be reached, and even higher than with known domains in some cases.

WebTerm Connector is a plugin for STAR MT Translate which combines machine translation with validated terminology information. The aim is to provide “understandable” information in the target language using corporate language and terminology.

Research on speaker-adapted neural machine translation (NMT) is scarce. One of the main challenges for more personalized MT systems is finding large enough annotated parallel datasets with speaker information. Rabinovich et al. (2017) published an annotated parallel dataset for EN–FR and EN–DE, however, for many other language pairs no sufficiently large annotated datasets are available.

Given (i) the rise of a new paradigm to machine translation based on neural networks that results in more fluent and less literal output than previous models and (ii) the maturity of machine-assisted translation via post-editing in industry, project PiPeNovel studies the feasibility of the post-editing workflow for literary text conducting experiments with professional literary translators.

Thanks to the great progress seen in the machine translation (MT) field in recent years, the use and perception of MT by translators need to be revisited. The main objective of this paper is to determine the perception, productivity and the post-editing effort (in terms of time and number of editings) of six translators when using Statistical Machine Translation (SMT) and Neural Machine Translation (NMT) systems. This presentation is focused on how translators perceive these two systems in order to know which one they prefer and what type of errors and problems present each system, as well as how translators solve these issues. These tests will be performed with the Dynamic Quality Framework (DQF) tools (quick comparison and productivity tasks) using Google Neural Machine Translation and Microsoft Translator (SMT) APIs in two different English into Spanish texts, an instruction manual and a marketing webpage. Results showed that translators considerably prefer NMT over SMT. Moreover, NMT is more adequate and fluent than SMT.

We test a series of techniques to predict punctuation and its effect on machine translation (MT) quality. Several techniques for punctuation prediction are compared: language modeling techniques, such as n-grams and long short-term memories (LSTM), sequence labeling LSTMs (unidirectional and bidirectional), and monolingual phrase-based, hierarchical and neural MT. For actual translation, phrase-based, hierarchical and neural MT are investigated. We observe that for punctuation prediction, phrase-based statistical MT and neural MT reach similar results, and are best used as a preprocessing step which is followed by neural MT to perform the actual translation. Implicit punctuation insertion by a dedicated neural MT system, trained on unpunctuated source and punctuated target, yields similar results.

In this paper, we present the challenge entailed by implementing a mobile version of a neural machine translation system, where the goal is to maximise translation quality while minimising model size. We explain the whole process of implementing the translation engine on an English–Spanish example and we describe all the difficulties found and the solutions implemented. The main techniques used in this work are data selection by means of Infrequent n-gram Recovery, appending a special word at the end of each sentence, and generating additional samples without the final punctuation marks. The last two techniques were devised with the purpose of achieving a translation model that generates sentences without the final full stop, or other punctuation marks. Also, in this work, the Infrequent n-gram Recovery was used for the first time to create a new corpus, and not enlarge the in-domain dataset. Finally, we get a small size model with quality good enough to serve for daily use.

According to Torres Hostench et al. (2016), the use of machine translation (MT) in Catalan and Spanish translation companies is low. Based on these results, the Tradumàtica research group, through the ProjecTA and ProjecTA-U projects, set to bring MT and translators closer with a two-fold strategy. On the one hand, by developing MTradumàtica, a free Moses-based web platform with graphical user interface (GUI) for statistical machine translation (SMT) trainers. On the other hand, by including MT-related contents in translators’ training. This paper will describe the latest developments in MTradumàtica.

After many successful experiments it has become evident that smart pre- and post-processing can significantly improve the output of neural machine translation. Therefore, various generic and language-specific processes are applied to the training corpus, the user input and the MT output for STAR MT Translate.

We present mtrain, a convenience tool for machine translation. It wraps existing machine translation libraries and scripts to ease their use. mtrain is written purely in Python 3, well-documented, and freely available.

This paper reports the results of two studies carried out with two different group of professional translators to find out how professionals perceive and accept SMT in comparison with TM. The first group translated and post-edited segments from English into German, and the second group from English into Spanish. Both studies had equivalent settings in order to guarantee the comparability of the results. It will also help to shed light upon the real benefit of SMT from which translators may take advantage.

Bidaide is a web service that allows the visitors of a museum, route or building to read or listen to explanations relative to the visited place on their own mobile and in their own language. The visitor can access the explanations in various ways: by scanning some QR codes located in the place, by GPS positioning (in outdoor routes), or by automatic Bluetooth proximity activation. This makes it accessible for people with reduced or null vision. On the other hand, this platform also offers to the manager of the visited site the most advanced language resources to create the texts and audios of the explanations in many languages.

We present the latest version of the SUMMA platform, an open-source software platform for monitoring and interpreting multi-lingual media, from written news published on the internet to live media broadcasts via satellite or internet streaming.

The requirement for neural machine translation (NMT) models to use fixed-size input and output vocabularies plays an important role for their accuracy and generalization capability. The conventional approach to cope with this limitation is performing translation based on a vocabulary of sub-word units that are predicted using statistical word segmentation methods. However, these methods have recently shown to be prone to morphological errors, which lead to inaccurate translations. In this paper, we extend the source-language embedding layer of the NMT model with a bi-directional recurrent neural network that generates compositional representations of the source words from embeddings of character n-grams. Our model consistently outperforms conventional NMT with sub-word units on four translation directions with varying degrees of morphological complexity and data sparseness on the source side.

While systems using the Neural Network-based Machine Translation (NMT) paradigm achieve the highest scores on recent shared tasks, phrase-based (PBMT) systems, rule-based (RBMT) systems and other systems may get better results for individual examples. Therefore, combined systems should achieve the best results for MT, particularly if the system combination method can take advantage of the strengths of each paradigm. In this paper, we describe a system that predicts whether a NMT, PBMT or RBMT will get the best Spanish translation result for a particular English sentence in DGT-TM 20161. Then we use fuzzy-match repair (FMR) as a mechanism to show that the combined system outperforms individual systems in a black-box machine translation setting.

We address the issues arising when a neural machine translation engine trained on generic data receives requests from a new domain that contains many specific technical terms. Given training data of the new domain, we consider two alternative methods to adapt the generic system: corpus-based and instance-based adaptation. While the first approach is computationally more intensive in generating a domain-customized network, the latter operates more efficiently at translation time and can handle on-the-fly adaptation to multiple domains. Besides evaluating the generic and the adapted networks with conventional translation quality metrics, in this paper we focus on their ability to properly handle domain-specific terms. We show that instance-based adaptation, by fine-tuning the model on-the-fly, is capable to significantly boost the accuracy of translated terms, producing translations of quality comparable to the expensive corpus-based method.

Automatic machine translation evaluation was crucial for the rapid development of machine translation systems over the last two decades. So far, most attention has been paid to the evaluation metrics that work with text on the sentence level and so did the translation systems. Across-sentence translation quality depends on discourse phenomena that may not manifest at all when staying within sentence boundaries (e.g. coreference, discourse connectives, verb tense sequence etc.). To tackle this, we propose several document-level MT evaluation metrics: generalizations of sentence-level metrics, language-(pair)-independent versions of lexical cohesion scores and coreference and morphology preservation in the target texts. We measure their agreement with human judgment on a newly created dataset of pairwise paragraph comparisons for four language pairs.

We present a comparison of automatic metrics against human evaluations of translation quality in several scenarios which were unexplored up to now. Our experimentation was conducted on translation hypotheses that were problematic for the automatic metrics, as the results greatly diverged from one metric to another. We also compared three different translation technologies. Our evaluation shows that in most cases, the metrics capture the human criteria. However, we face failures of the automatic metrics when applied to some domains and systems. Interestingly, we find that automatic metrics applied to the neural machine translation hypotheses provide the most reliable results. Finally, we provide some advice when dealing with these problematic domains.

This paper presents a shallow-transfer machine translation (MT) system for translating from Kazakh to Turkish. Background on the differences between the languages is presented, followed by how the system was designed to handle some of these differences. The system is based on the Apertium free/open-source machine translation platform. The structure of the system and how it works is described, along with an evaluation against two competing systems. Linguistic components were developed, including a Kazakh-Turkish bilingual dictionary, Constraint Grammar disambiguation rules, lexical selection rules, and structural transfer rules. With many known issues yet to be addressed, our RBMT system has reached performance comparable to publicly-available corpus-based MT systems between the languages.

Recurrent neural networks (RNNs) have represented for years the state of the art in neural machine translation. Recently, new architectures have been proposed, which can leverage parallel computation on GPUs better than classical RNNs. Faster training and inference combined with different sequence-to-sequence modeling also lead to performance improvements. While the new models completely depart from the original recurrent architecture, we decided to investigate how to make RNNs more efficient. In this work, we propose a new recurrent NMT architecture, called Simple Recurrent NMT, built on a class of fast and weakly-recurrent units that use layer normalization and multiple attentions. Our experiments on the WMT14 English-to-German and WMT16 English-Romanian benchmarks show that our model represents a valid alternative to LSTMs, as it can achieve better results at a significantly lower computational cost.

We propose and study three different novel approaches for tackling the problem of development set selection in Statistical Machine Translation. We focus on a scenario where a machine translation system is leveraged for translating a specific test set, without further data from the domain at hand. Such test set stems from a real application of machine translation, where the texts of a specific e-commerce were to be translated. For developing our development-set selection techniques, we first conducted experiments in a controlled scenario, where labelled data from different domains was available, and evaluated the techniques both with classification and translation quality metrics. Then, the best-performing techniques were evaluated on the e-commerce data at hand, yielding consistent improvements across two language directions.

A large percentage of the world’s population speaks a language of the Indian subcontinent, what we will call here Indic languages, comprising languages from both Indo-European (e.g., Hindi, Bangla, Gujarati, etc.) and Dravidian (e.g., Tamil, Telugu, Malayalam, etc.) families, upwards of 1.5 Billion people. A universal characteristic of Indic languages is their complex morphology, which, when combined with the general lack of sufficient quantities of high quality parallel data, can make developing machine translation (MT) for these languages difficult. In this paper, we describe our efforts towards developing general domain English–Bangla MT systems which are deployable to the Web. We initially developed and deployed SMT-based systems, but over time migrated to NMT-based systems. Our initial SMT-based systems had reasonably good BLEU scores, however, using NMT systems, we have gained significant improvement over SMT baselines. This is achieved using a number of ideas to boost the data store and counter data sparsity: crowd translation of intelligently selected monolingual data (throughput enhanced by an IME (Input Method Editor) designed specifically for QWERTY keyboard entry for Devanagari scripted languages), back-translation, different regularization techniques, dataset augmentation and early stopping.

Translation Quality Estimation (QE) aims to estimate the quality of an automated machine translation (MT) output without any human intervention or reference translation. With the increasing use of MT systems in various cross-lingual applications, the need and applicability of QE systems is increasing. We study existing approaches and propose multiple neural network approaches for sentence-level QE, with a focus on MT outputs in Indian languages. For this, we also introduce five new datasets for four language pairs: two for English–Gujarati, and one each for English–Hindi, English–Telugu and English–Bengali, which includes one manually post-edited dataset for English–Gujarati. These Indian languages are spoken by around 689M speakers world-wide. We compare results obtained using our proposed models with multiple state-of-the-art systems including the winning system in the WMT17 shared task on QE and show that our proposed neural model which combines the discriminative power of carefully chosen features with Siamese Convolutional Neural Networks (CNNs) works best for all Indian language datasets.

I will try to answer the question of whether Machine Translation (MT) can be considered a full translation process. I argue that, instead, it should be seen as part of a process performed by translators, in which MT plays a fundamental support role. The roles of translators and MT in the translation process is presented in an analysis that get its elements from Translation Studies and Translation Process Research.

While MT+PE has become an industry standard, our translation schools are not able to accompany these changes by updating their academic programs. We polled 100 pre-professionals to confirm that in our local context they are reluctant to accept post-editing jobs mainly because they have inherited pre-conceptions or negative opinions about MT during their studies.

The annual WMT Conference in Machine Translation has been running shared tasks since 2006. It started with a translation task based on Europarl, and has grown to include tasks on all aspects of MT corpus preparation, training and evaluation, including the flagship task on news translation. I will review the history of the task, lessons learnt, and plans for future tasks.

Translation memories (TM) and machine translation (MT) both are potentially useful resources for professional translators, but they are often still used independently in translation workflows. As translators tend to have a higher confidence in fuzzy matches than in MT, we investigate how to combine the benefits of TM retrieval with those of MT, by integrating the results of both. We develop a flexible TM-MT integration approach based on various techniques combining the use of TM and MT, such as fuzzy repair, span pretranslation and exploiting multiple matches. Results for ten language pairs using the DGT-TM dataset indicate almost consistently better BLEU, METEOR and TER scores compared to the MT, TM and NMT baselines.

Previous attempts at injecting semantic frame biases into SMT training for low resource languages failed because either (a) no semantic parser is available for the low resource input language; or (b) the output English language semantic parses excise relevant parts of the alignment space too aggressively. We present the first semantic SMT model to succeed in significantly improving translation quality across many low resource input languages for which no automatic SRL is available —consistently and across all common MT metrics. The results we report are the best by far to date for this type of approach; our analyses suggest that in general, easier approaches toward including semantics in training SMT models may be more feasible than generally assumed even for low resource languages where semantic parsers remain scarce. While recent proposals to use the crosslingual evaluation metric XMEANT during inversion transduction grammar (ITG) induction are inapplicable to low resource languages that lack semantic parsers, we break the bottleneck via a vastly improved method of biasing ITG induction toward learning more semantically correct alignments using the monolingual semantic evaluation metric MEANT. Unlike XMEANT, MEANT requires only a readily-available English (output language) semantic parser. The advances we report here exploit the novel realization that MEANT represents an excellent way to semantically bias expectation-maximization induction even for low resource languages. We test our systems on challenging languages including Amharic, Uyghur, Tigrinya and Oromo. Results show that our model influences the learning towards more semantically correct alignments, leading to better translation quality than both the standard ITG or GIZA++ based SMT training models on different datasets.

This product presentation describes the integration of the three MT technologies currently used – rule-based (RBMT), Statistical (SMT) and Neural (NMT) – into one scalable single platform, OctaveMT. MT clients can access all three types of MT engines, whether on a user specified basis or depending on several translation parameters (language-direction, domain, etc.)

The advances of deep learning approaches in automatic speech recognition (ASR) and machine translation (MT) have allowed for levels of accuracy that move speech translation closer to being a commercially viable alternative interpretation solution. In addition, recent improvements in micro-electronic mechanical systems, microphone arrays, speech processing software, and wireless technology have enabled speech recognition software to capture higher quality speech input from wireless earpiece products. With this in mind, we introduce and present a wearable speech translation tool called Pilot, which uses these systems to translate language spoken within the proximity of a user wearing the wireless earpiece.

The overall machine translation quality available for professional translators working with the Spanish–Basque pair is rather poor, which is a deterrent for its adoption. This work investigates the plausibility of building a comprehensive recommendation system to speed up decision time between post-editing or translation from scratch using the very limited training data available. First, we build a set of regression models that predict the post-editing effort in terms of overall quality, time and edits. Secondly, we build classification models that recommend the most efficient editing approach using post-editing effort features on top of linguistic features. Results show high correlations between the predictions of the regression models and the expected HTER, time and edit number values. Similarly, the results for the classifiers show that they are able to predict with high accuracy whether it is more efficient to translate or to post-edit a new segment.

A salient feature of Neural Machine Translation (NMT) is the end-to-end nature of training employed, eschewing the need of separate components to model different linguistic phenomena. Rather, an NMT model learns to translate individual sentences from the labeled data itself. However, traditional NMT methods trained on large parallel corpora with a one-to-one sentence mapping make an implicit assumption of sentence independence. This makes it challenging for current NMT systems to model inter-sentential discourse phenomena. While recent research in this direction mainly leverages a single previous source sentence to model discourse, this paper proposes the incorporation of a context window spanning previous as well as next sentences as source-side context and previously generated output as target-side context, using an effective non-recurrent architecture based on self-attention. Experiments show improvement over non-contextual models as well as contextual methods using only previous context.

This paper presents the preliminary results of an ongoing academia-industry collaboration that aims to integrate MT into the workflow of Swiss Post’s Language Service. We describe the evaluations carried out to select an MT tool (commercial or open-source) and assess the suitability of machine translation for post-editing in Swiss Post’s various subject areas and language pairs. The goal of this first phase is to provide recommendations with regard to the tool, language pair and most suitable domain for implementing MT.