Invited Speakers

We have recently started to hear discussions about AI-driven “augmented translation”, with suggestions that this will represent a “profound technological shift that will affect all parties” (Lommel and DePalma, 2021). In this talk I will examine what is generally meant by “augmented translation”, whether this is a current trend, or a trend we can expect in the (near?) future. Or, is it just a trendy term for reconceptualising technology that already exists? Examining how “augmentation” has been viewed in the past and how it is being discussed and implemented in other areas will help us to consider its possibilities and limitations for translation. Finally, I will focus on the feasibility and desirability of “augmentation”, asking who might actually benefit from it, if it were achievable.

Lommel, Arle and Don DePalma (2021) “Augmented Translation: How Artificial Intelligence Drives Productivity and Efficiency for the Language Industry”. Common Sense Advisory Research.

Machine translation relies heavily on data. In rule-based machine translation, an engine performs the translation task using language resources such as dictionaries and grammar rules, usually written by experts, but sometimes learned from monolingual or bilingual texts. Automatic corpus-based (statistical and, more recently, neural) machine translation takes advantage of large amounts of bilingual (sentence-aligned) and monolingual texts, very often harvested using web-crawling techniques. Clearly, the machine translation software that exploits these data to translate is creative work that may be copyrighted, but this talk will focus on the data instead. Human work, and therefore the creative authorship of works, is present in all forms of machine translation data: monolingual text has been written, parallel text has been translated and aligned, and rules and dictionaries have been written by experts. Since it was conceived centuries ago, copyright protects the livelihoods of authors by regulating how copies of these data may be used and how the resulting works may be used and published, through instruments such as licences. Although the case for dictionaries and grammars as used in rule-based machine translation is reasonably clear, as they are written specifically for one language processing application or other, monolingual and parallel text, as used in machine translation, was not created with machine translation in mind, and this has led some authors to wonder whether authors and translators should obtain additional compensation for this unexpected use of their work to generate new “downstream” value. The conference provides an overview of the different data sources used in machine translation, discusses their authorship throughout the steps of creation, care and transformation of this data for use with machine translation, and touches on the types of implicit and explicit licensing schemes that apply to them and how they work. I will also discuss the controversy surrounding the use of published works to generate new, initially unforeseen value through translation technologies, and the different ways in which copyright issues are addressed in practice.

Machine translation (MT) is typically formulated as the task of returning a correct translation for a single input sentence, and as a result, many alternative translations of a sentence can be considered as valid. However, human translations are mostly generated and consumed for specific communicative purposes and situations. Both verbal and social context provide important clues for generating appropriate translations, that are not only correct in terms of content but that also address the audience with the correct style and register. Thus an ideal MT system would be able to control language variations via context, such as the previous sentences, the characteristics of the speaker, the relationship between the speaker and the audience, the domain of discourse, the specific use case, and so on. The recent “neural revolution” brought us impressive improvements in MT quality but also a new paradigm to approach AI problems. However, if we look at how current state-of-the-art MT works in communicative situations where context plays an important role, we get the impression that we still have a long journey ahead. In this talk, I will present results of work carried out at AWS AI Labs to close this gap and enhance neural MT along different linguistic traits that depend on context information: discourse domain, translation style, verbosity and register. I will also discuss a challenging use case in which contextual information is extremely important: the translation of movie scripts.

In recent decades, rapid advances in language technology have been leaving a heavy imprint on the way legal translation is organized in both the private and public sectors.

The European Commission’s Directorate-General for Translation has been at the forefront of digital transformation in the translation profession in the public sector, introducing and developing eTranslation, its machine translation system, sponsored by the Connecting Europe Facility.

In her presentation Merit-Ene Ilja focuses on the language technology applications and tools used in the DGT’s translation ecosystem to support the human-in-the-loop translation workflow. Language technology is seen as a key component in DGT’s resource mix to enable effective and efficient translation of EU legislation and communication material. The use of benefits offered by technology is made possible thanks to staff and continued investment in their professional development. Today, DGT’s learning and development priorities for translation staff are centred around enhancing their digital proficiency, thematic knowledge and data management skills, vital for quality and efficiency in translation work.

In my presentation I will illustrate how the translator in the European Parliament has evolved over the last years: from translator to a versatile language professional. The main purpose of our language professionals is to provide multilingual products and to communicate clearly with multiple audiences in order to foster a better understanding of the work of the European Parliament and its Members. Increasingly, the focus is on three formats: text, audio and video. DG TRAD continues to excel as a global leader in the translation field, thanks to the work of our language professionals and innovative technology