This paper presents the University of Edinburgh’s constrained submissions of English-German and English-Hausa systems to the WMT 2021 shared task on news translation. We build En-De systems in three stages: corpus filtering, back-translation, and fine-tuning. For En-Ha we use an iterative back-translation approach on top of pre-trained En-De models and investigate vocabulary embedding mapping.

We describe our submission to the IWSLT 2021 shared task on simultaneous text-to-text English-German translation. Our system is based on the re-translation approach where the agent re-translates the whole source prefix each time it receives a new source token. This approach has the advantage of being able to use a standard neural machine translation (NMT) inference engine with beam search, however, there is a risk that incompatibility between successive re-translations will degrade the output. To improve the quality of the translations, we experiment with various approaches: we use a fixed size wait at the beginning of the sentence, we use a language model score to detect translatable units, and we apply dynamic masking to determine when the translation is unstable. We find that a combination of dynamic masking and language model score obtains the best latency-quality trade-off.

Europe is a multilingual society, in which dozens of languages are spoken. The only option to enable and to benefit from multilingualism is through Language Technologies (LT), i.e., Natural Language Processing and Speech Technologies. We describe the European Language Grid (ELG), which is targeted to evolve into the primary platform and marketplace for LT in Europe by providing one umbrella platform for the European LT landscape, including research and industry, enabling all stakeholders to upload, share and distribute their services, products and resources. At the end of our EU project, which will establish a legal entity in 2022, the ELG will provide access to approx. 1300 services for all European languages as well as thousands of data sets.

With 24 official EU and many additional languages, multilingualism in Europe and an inclusive Digital Single Market can only be enabled through Language Technologies (LTs). European LT business is dominated by hundreds of SMEs and a few large players. Many are world-class, with technologies that outperform the global players. However, European LT business is also fragmented – by nation states, languages, verticals and sectors, significantly holding back its impact. The European Language Grid (ELG) project addresses this fragmentation by establishing the ELG as the primary platform for LT in Europe. The ELG is a scalable cloud platform, providing, in an easy-to-integrate way, access to hundreds of commercial and non-commercial LTs for all European languages, including running tools and services as well as data sets and resources. Once fully operational, it will enable the commercial and non-commercial European LT community to deposit and upload their technologies and data sets into the ELG, to deploy them through the grid, and to connect with other resources. The ELG will boost the Multilingual Digital Single Market towards a thriving European LT community, creating new jobs and opportunities. Furthermore, the ELG project organises two open calls for up to 20 pilot projects. It also sets up 32 national competence centres and the European LT Council for outreach and coordination purposes.

This paper describes the University of Edinburgh’s neural machine translation systems submitted to the IWSLT 2020 open domain Japanese↔Chinese translation task. On top of commonplace techniques like tokenisation and corpus cleaning, we explore character mapping and unsupervised decoding-time adaptation. Our techniques focus on leveraging the provided data, and we show the positive impact of each technique through the gradual improvement of BLEU.

We describe the joint submission of the University of Edinburgh and Charles University, Prague, to the Czech/English track in the WMT 2020 Shared Task on News Translation. Our fast and compact student models distill knowledge from a larger, slower teacher. They are designed to offer a good trade-off between translation quality and inference efficiency. On the WMT 2020 Czech ↔ English test sets, they achieve translation speeds of over 700 whitespace-delimited source words per second on a single CPU thread, thus making neural translation feasible on consumer hardware without a GPU.

This paper describes the University of Edinburgh’s submission of German <-> English systems to the WMT2020 Shared Tasks on News Translation and Zero-shot Robustness.

The University of Edinburgh participated in the WMT19 Shared Task on News Translation in six language directions: English↔Gujarati, English↔Chinese, German→English, and English→Czech. For all translation directions, we created or used back-translations of monolingual data in the target language as additional synthetic training data. For English↔Gujarati, we also explored semi-supervised MT with cross-lingual language model pre-training, and translation pivoting through Hindi. For translation to and from Chinese, we investigated character-based tokenisation vs. sub-word segmentation of Chinese text. For German→English, we studied the impact of vast amounts of back-translated training data on translation quality, gaining a few additional insights over Edunov et al. (2018). For English→Czech, we compared different preprocessing and tokenisation regimes.

The open-source SUMMA Platform is a highly scalable distributed architecture for monitoring a large number of media broadcasts in parallel, with a lag behind actual broadcast time of at most a few minutes. It assembles numerous state-of-the-art NLP technologies into a fully automated media ingestion pipeline that can record live broadcasts, detect and transcribe spoken content, translate from several languages (original text or transcribed speech) into English, recognize Named Entities, detect topics, cluster and summarize documents across language barriers, and extract and store factual claims in these news items. This paper describes the intended use cases and discusses the system design decisions that allowed us to integrate state-of-the-art NLP modules into an effective workflow with comparatively little effort.

The University of Edinburgh made submissions to all 14 language pairs in the news translation task, with strong performances in most pairs. We introduce new RNN-variant, mixed RNN/Transformer ensembles, data selection and weighting, and extensions to back-translation.

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 open-source SUMMA Platform is a highly scalable distributed architecture for monitoring a large number of media broadcasts in parallel, with a lag behind actual broadcast time of at most a few minutes. The Platform offers a fully automated media ingestion pipeline capable of recording live broadcasts, detection and transcription of spoken content, translation of all text (original or transcribed) into English, recognition and linking of Named Entities, topic detection, clustering and cross-lingual multi-document summarization of related media items, and last but not least, extraction and storage of factual claims in these news items. Browser-based graphical user interfaces provide humans with aggregated information as well as structured access to individual news items stored in the Platform’s database. This paper describes the intended use cases and provides an overview over the system’s implementation.

We present Marian, an efficient and self-contained Neural Machine Translation framework with an integrated automatic differentiation engine based on dynamic computation graphs. Marian is written entirely in C++. We describe the design of the encoder-decoder framework and demonstrate that a research-friendly toolkit can achieve high training and translation speed.

We present the first prototype of the SUMMA Platform: an integrated platform for multilingual media monitoring. The platform contains a rich suite of low-level and high-level natural language processing technologies: automatic speech recognition of broadcast media, machine translation, automated tagging and classification of named entities, semantic parsing to detect relationships between entities, and automatic construction / augmentation of factual knowledge bases. Implemented on the Docker platform, it can easily be deployed, customised, and scaled to large volumes of incoming media streams.

We investigate techniques for supervised domain adaptation for neural machine translation where an existing model trained on a large out-of-domain dataset is adapted to a small in-domain dataset. In this scenario, overfitting is a major challenge. We investigate a number of techniques to reduce overfitting and improve transfer learning, including regularization techniques such as dropout and L2-regularization towards an out-of-domain prior. In addition, we introduce tuneout, a novel regularization technique inspired by dropout. We apply these techniques, alone and in combination, to neural machine translation, obtaining improvements on IWSLT datasets for English→German and English→Russian. We also investigate the amounts of in-domain training data needed for domain adaptation in NMT, and find a logarithmic relationship between the amount of training data and gain in BLEU score.

This paper presents a phrase table implementation for the Moses system that computes phrase table entries for phrase-based statistical machine translation (PBSMT) on demand by sampling an indexed bitext. While this approach has been used for years in hierarchical phrase-based translation, the PBSMT community has been slow to adopt this paradigm, due to concerns that this would be slow and lead to lower translation quality. The experiments conducted in the course of this work provide evidence to the contrary: without loss in translation quality, the sampling phrase table ranks second out of four in terms of speed, being slightly slower than hash table look-up (Junczys-Dowmunt, 2012) and considerably faster than current implementations of the approach suggested by Zens and Ney (2007). In addition, the underlying parallel corpus can be updated in real time, so that professionally produced translations can be used to improve the quality of the machine translation engine immediately.

We present a rule-based, deterministic dependency parser for Japanese. It was implemented in C++, using object classes that reflect linguistic concepts and thus facilitate the transfer of linguistic intuitions into code. The parser first chunks morphemes into one-word phrases and then parses from the right to the left. The average parsing accuracy is 83.6%.

We present an approach to semantic interpretation of syntactically parsed Japanese sentences that works largely parser-independent. The approach relies on a standardized parse tree format that restricts the number of syntactic configurations that the semantic interpretation rules have to anticipate. All parse trees are converted to this format prior to semantic interpretation. This setup allows us not only to apply the same set of semantic interpretation rules to output from different parsers, but also to independently develop parsers and semantic interpretation rules.