Language tagging, a method whereby source and target inputs are prefixed with a unique language token, has become the de facto standard for conditioning Multilingual Neural Machine Translation (MNMT) models on specific language directions. This conditioning can manifest effective zero-shot translation abilities in MT models at scale for many languages. Expanding on previous work, we propose a novel method of language tagging for MNMT, injection, in which the embedded representation of a language token is concatenated to the input of every linear layer. We explore a variety of different tagging methods, with and without injection, showing that injection improves zero-shot translation performance with up to a 2+ BLEU score point gain for certain language directions in our dataset.

The morphological structure of Semitic languages, such as Arabic, is based on non-concatenative roots and templates. This complex word structure used by humans is obscured to neural models that employ traditional tokenization algorithms, such as byte-pair encoding (BPE) (Sennrich et al., 2016; Gage, 1994). In this work, we present and evaluate Semitic Root Encoding (SRE), a tokenization method that represents both concatenative and non-concatenative structures in Semitic words with sequences of root, template stem, and BPE tokens. We apply the method to neural machine translation (NMT) and find that SRE tokenization yields an average increase of 1.15 BLEU over the baseline. SRE tokenization is also robust against generating combinations of roots with template stems that do not occur in nature. Finally, we compare the performance of SRE to tokenization based on non-linguistic root and template structures and tokenization based on stems, providing evidence that NMT models are capable of leveraging tokens based on non-concatenative Semitic morphology.

Multilingual Neural Machine Translation (MNMT) models enhance translation quality for low-resource languages by exploiting cross-lingual similarities during training—a process known as knowledge transfer. This transfer is particularly effective between languages that share lexical or structural features, often enabled by a common orthography. However, languages with strong phonetic and lexical similarities but distinct writing systems experience limited benefits, as the absence of a shared orthography hinders knowledge transfer. To address this limitation, we propose an approach based on phonetic information that enhances token-level alignment across scripts by leveraging transliterations. We systematically evaluate several phonetic transcription techniques and strategies for incorporating phonetic information into NMT models. Our results show that using a shared encoder to process orthographic and phonetic inputs separately consistently yields the best performance for Khmer, Thai, and Lao in both directions with English, and that our custom Cognate-Aware Transliteration (CAT) method consistently improves translation quality over the baseline.

We propose an approach that improves the performance of VMT (Video-guided Machine Translation) models, which integrate text and video modalities. We experiment with the MAD (Movie Audio Descriptions) dataset, a new dataset which contains transcribed audio descriptions of movies. We find that the MAD dataset is more lexically rich than the VATEX dataset (the current VMT baseline), and we experiment with MAD pretraining to improve performance on the VATEX dataset. We experiment with two different video encoder architectures: a Conformer (Convolution-augmented Transformer) and a Transformer. Additionally, we conduct experiments by masking the source sentences to assess the degree to which the performance of both architectures improves due to pretraining on additional video data. Finally, we conduct an analysis of the transfer learning potential of a video dataset and compare it to pretraining on a text-only dataset. Our findings demonstrate that pretraining with a lexically rich dataset leads to significant improvements in model performance when models use both text and video modalities.

Lingua is an application developed for the Church of Jesus Christ of Latter-day Saints that performs both real-time interpretation of live speeches and automatic video dubbing (AVD). Like other AVD systems, it can perform synchronized automatic dubbing, given video files and optionally, corresponding text files using a traditional ASR–MT–TTS pipeline. Lingua’s unique contribution is that it can also operate in real-time with a slight delay of a few seconds to interpret live speeches. If no source-language script is provided, the translations are exactly as recognized by ASR and translated by MT. If a script is provided, Lingua matches the recognized ASR segments with script segments and passes the latter to MT for translation and subsequent TTS. If a human translation is also provided, it is passed directly to TTS. Lingua switches between these modes dynamically, enabling translation of off-script comments and different levels of quality for multiple languages. (see extended abstract)

The Church of Jesus Christ of Latter-day Saints undertook an extensive effort at the beginning of this year to deploy machine translation (MT) in the translation workflow for the content on its principal website, www.lds.org. The objective of this effort is to reduce by at least 50% the time required by human translators to translate English content into nine other languages and publish it on this site. This paper documents the experience to date, including selection of the MT system, preparation and use of data to customize the system, initial deployment of the system in the Church’s translation workflow, post-editing training for translators, the resulting productivity improvements, and plans for future deployments.

At AMTA 2002, we reported on a pilot project to machine translate Microsoft’s Product Support Knowledge Base into Spanish. The successful pilot has since resulted in the permanent deployment of both Spanish and Japanese versions of the knowledge base, as well as ongoing pilot projects for French and German. The translated articles in each case have been produced by MSR-MT, Microsoft Research’s data-driven MT system, which has been trained on well over a million bilingual sentence pairs for each target language from previously translated materials contained in translation memories and glossaries. This paper describes our experience in deploying this system and the (positive) customer response to the availability of machine translated articles, as well as other uses of MSR-MT either planned or underway at Microsoft.

MSR-MT is an advanced research MT prototype that combines rule-based and statistical techniques with example-based transfer. This hybrid, large-scale system is capable of learning all its knowledge of lexical and phrasal translations directly from data. MSR-MT has undergone rigorous evaluation showing that, trained on a corpus of technical data similar to the test corpus, its output surpasses the quality of best-of-breed commercial MT systems.

Translation systems that automatically extract transfer mappings (rules or examples) from bilingual corpora have been hampered by the difficulty of achieving accurate alignment and acquiring high quality mappings. We describe an algorithm that uses a best-first strategy and a small alignment grammar to significantly improve the quality of the mappings extracted. For each mapping, frequencies are computed and sufficient context is retained to distinguish competing mappings during translation. Variants of the algorithm are run against a corpus containing 200K sentence pairs and evaluated based on the quality of resulting translations.