Machine translation for everyone

Front matter

Machine translation for everyone

Empowering users in the age of artificial intelligence

Edited by

Contents

Acknowledgments iii Introduction Dorothy Kenny v

1 Europe, multilingualism and machine translation Olga Torres-Hostench 1

2 Human and machine translation Dorothy Kenny 23

Introduction

Dorothy Kenny Dublin City University

In this Introduction I set out the rationale for this book and suggest ways in which readers might approach the material it contains.

1 Why this book?

Multilingualism is a foundational value of the European Union. In practice, it depends on multiple pillars, including language learning and translation. In recent years, both of these pillars have been profoundly affected by the continued development of machine translation. Machine translation, if used wisely, can serve to support language learning efforts. It can provide access to more texts, originally written in more languages, to more users than ever before. And it can help professional human translators to work more productively, given the right circumstances. There is a danger, however, that uncritical use of the technology could set back individual efforts to learn languages. It could also be used in ways that gloss over potential problems with the output it produces, so that consumers don't realize when texts they are reading contain machine-translation induced inaccuracies or biases. And if excessively hyped, it could discourage new blood from entering the translation profession.

In order to harness the benefits of machine translation without jeopardizing other valuable pillars of multilingualism, we take the view, following Bowker & Ciro (2019), that good multilingual citizenship and professional translation must be underpinned by machine translation literacy. Different levels of machine translation literacy will be appropriate for different people. Bowker & Ciro (2019) are concerned with the international research community, for example. Here we target two main constituencies: more “occasional” users of machine translation - those who use the technology for ad hoc information gathering purposes, or even unwittingly when they consume machine translated text without realizing it, or tion for everyone: Empowering users in the age of artificial intelligence, v—viii.

N | Dorothy Kenny. 2022. Introduction. In Dorothy Kenny (ed.), Machine transla- | Berlin: Language Science Press.

casually in language learning contexts; and people who are either already working as translators or training to become translators. We take the view that all users of machine translation should have some basic understanding of why the technology is important, and where it fits into the maintenance of multilingual regimes. And all users should have some basic understanding of how the technology works, so they can use it intelligently and avoid common pitfalls. Some users, who may wish to engage more deeply with the technology, may benefit from knowing how to get the best out of machine translation, for example, by writing texts in a way that makes them easier to translate by machine. The same users might also be interested in ways to improve machine translation outputs. Those working in, or about to join, the translation industry, will have a particular interest in evaluating machine translation output, in order to gauge whether it is “fit for purpose". They might even get involved in integrating machine translation into the workflow of their company or need to know how to customize machine translation so that they can better serve the needs of particular clients. They will also be interested in how machine translation might impact on their working conditions. Such readers require more in-depth knowledge of the technology itself, and of the techniques and tools they can use to implement it. All users should have some basic knowledge of the ethical issues that arise when we use machine translation, for different reasons. Some users may be concerned about the possibility of cheating: in what cases might the use of machine translation constitute a breach of trust in educational environments, for example? Others, mainly professional translators, may have to consider how the use of certain types of machine translation might constitute a breach of contract. And everybody has to be concerned these days about protecting the privacy and data rights of others. Contemporary machine translation is also one of the many technologies that can be implicated in processes that degrade our natural environment. And it has been known to produce biased outputs, preferring male to female forms, for example. Like all communication technologies, it can be used for nefarious causes or positive humanitarian purposes. These are issues that concern us all.

2 How to use this book

In this book we attempt to guide readers through all of the above issues. We do not assume any prior knowledge of either translation in general or machine translation in particular. When we do move on to the more technical aspects of machine translation, and especially neural machine translation - a state-ofvi

Introduction the-art translation technology based on artificial neural networks — we first describe these technologies without recourse to mathematical concepts that may not be familiar to readers. We attempt to use insightful explanations, and especially metaphors, that will help readers understand the general concepts that inform the area. In so doing, we provide a gentle introduction to the contemporary world of machine learning. We gradually build up a rich picture of machine translation. In general, readers will find the earlier chapters less specialized than the later ones, and earlier sections of later chapters less specialized than subsequent sections. Some readers may be able to skip some chapters completely, but even machine translation specialists might benefit from reading “non-technical” chapters, for example, on machine translation and ethics.

3 The structure of this book

This book opens with Olga Torres-Hostench's discussion in Chapter 1 of multilingualism, what it means, and how it is operationalized, in particular in the European Union. She makes the case for the considered integration of machine translation into both language learning and translation. In Chapter 2, Dorothy Kenny discusses translation in general and machine translation in particular, aiming to dispel some myths about translation before gently introducing the reader to the basic concepts behind contemporary approaches to machine translation, including artificial intelligence and machine learning. Chapter 3, by Caroline Rossi and Alice Carré, brings us into the world of machine translation evaluation, an area of considerable scientific and economic importance. In Chapter 4, Pilar Sánchez- Gijón and Dorothy Kenny address the ways in which we can make translation easier for machines before the fact, while in Chapter 5, Sharon O’Brien guides us through approaches to after-the-fact improvement of machine translated texts through the activity known as post-editing. Chapter 6, by Joss Moorkens, discusses the many ethical issues that arise in contexts where machine translation is used. Chapter 7 is the most technical chapter in the book. In it, Juan Antonio Pérez-Ortiz, Mikel Forcada and Felipe Sánchez-Martínez explain how neural machine translation works, covering the basic techniques that are most commonly used in contemporary systems. Gema Ramírez-Sánchez's guide to custom neural machine translation follows in Chapter 8. The volume closes with a chapter dedicated to machine translation and language learning, written by Alice Carré, Dorothy Kenny, Caroline Rossi, Pilar Sánchez-Gijón and Olga Torres-Hostench.

vii

Accompanying resources

Each chapter of this book is accompanied by a set of interactive activities accessible through the MultiTraiNMT website at http://www.multitrainmt.eu/. A permanent link to these activities can be found at https://ddd.uab.cat/record/257869. Most of the activities can be completed on a self-access basis, although some will benefit from the guidance of a teacher.

A special pedagogical platform known as MutNMT has also been created as part of the MultiTraiNMT project. It is designed to help users learn how to train, customize and evaluate neural machine translation systems. It is accessible through the MultiTraiNMT website, and will be of particular significance to readers of Chapters 7 and 8 of this book.

Bowker, Lynne & Jairo Buitrago Ciro. 2019. Machine translation and global research. Bingley: Emerald Publishing.

viii

Chapter 1 Universitat Autónoma de Barcelona

This chapter explains multilingualism as a foundational principle of the European Union, describing how it is put into practice and supported through language learning and translation. Taking the university campus as a case study, it argues that machine translation can be used to foster multilingualism in this context.

The European Union's motto “united in diversity” is said to symbolize “the essential contribution that linguistic diversity and language learning make to the European project" (European Commission 2021). But European Union (EU) policy on multilingualism is mostly built upon language learning and mobility, both time-consuming activities. And human language learning presents particular challenges. After all, there is a limit to the number of languages the average EU citizen can learn. The aim in this chapter is to suggest answers to these questions by arguing that machine translation can contribute to the promotion of multilingualism in Europe and thus to European linguistic diversity.

2 A multilingual EU

It is ... an open secret that the EU's supposedly humane multilingualism is but an illusion. (House 2003: 561)

ISO 639-3 is a set of codes developed by the International Organization for Standardization (ISO) that defines three-letter identifiers for all known human

Olga Torres-Hostench. 2022. Europe, multilingualism and machine translation. In Dorothy Kenny (ed.), Machine translation for everyone: Empowering l users in the age of artificial intelligence, 1-21. Berlin: Language Science Press.

languages. As of 30 January 2020, the standard contained entries for 7,868 languages (Wikizero 2020), around 600 of which are spoken in Europe, and 24 of which are official languages of the EU. These are: Dutch, French, German, Italian (since 1958); Danish, English (since 1973); Greek (since 1981); Portuguese, Spanish (since 1986); Finnish, Swedish (since 1995); Czech, Estonian, Hungarian, Latvian, Lithuanian, Maltese, Polish, Slovak, Slovene (since 2004); Bulgarian, Irish, Romanian (since 2007) and Croatian (since 2013).

Linguistic diversity is part of Europe's cultural heritage. In Europe, there are languages with official status at state level, and indigenous regional and/or minority languages with different degrees of recognition. The 1998 European Charter for Regional or Minority Languages is the European convention for the protection and promotion of languages used by traditional minorities. It was reformed and strengthened by a monitoring mechanism in 2019. The Charter covers 79 languages used by 201 national minorities or linguistic groups (Council of Europe 2020). They are presented in alphabetical order in Table 1.

According to the Charter, some of these languages are to be protected in just one country, such as Skolt Sami in Finland, whereas others should be protected in several countries, such as Slovenian in Austria, Bosnia and Herzegovina, Croatia and Hungary. Beyond the Charter, there are other languages with different levels of recognition. For instance, Sardinia, an autonomous region of Italy, recognizes the Sardinian language as an official language, and Romansh Ladino, Cimbrian and Mocheno, spoken in certain communes of the mountainous North of Italy, also have local recognition.

The European Charter for Regional or Minority Languages, however, guarantees the rights only of regional minority groups, and not of migrant groups. What's more, the Charter has noteworthy absences, such as Breton, spoken in the North West of France, although a Breton language agency was created by the Region of Brittany in 2010 to promote daily use of the language.

Multilingualism in Europe is also enhanced by immigration and mobility. There have been intra-European migrations, leading, for example, to Portuguese being spoken in Andorra and Polish in Ireland, alongside languages traditionally spoken outside the EU, such as Mandarin Chinese or Arabic. In the Multilingual Cities Project (Extra & Yagmur 2005), home language surveys amongst pupils both in primary and secondary schools were collected in Brussels, Hamburg, Lyon, Madrid, The Hague and Góteborg. The list of collected languages was the following: Romani, Turkish, Urdu, Armenian, Russian, Serbian/Croatian/Bosnian, Albanian, Vietnamese, Chinese, Arabic, Polish, Somali, Portuguese, Berber, Kurdish, Spanish, French, Italian, English, German. The authors of the study reached an obvious but provocative conclusion:

Table 1: Languages covered by the European Charter for Regional or Minority Languages

Amongst the major 20 languages in the participating cities, 10 languages are of European origin and 10 languages stem from abroad. These findings show that the traditional concept of language diversity in Europe should be reconsidered and extended. (Extra & Yagmur 2005)

But what is the "traditional concept" of language diversity?

2.1 The 24 official languages as a symbol of European linguistic diversity

The European Union considers its linguistic diversity a valuable asset. Article 22 of the Charter of Fundamental Rights of the European Union (The Member States, 2012) states that “[t]he Union shall respect cultural, religious and linguistic diversity" However, member states have the exclusive right to define and recognize national and regional minority languages, and language policies are highly controversial.

Meanwhile, the EU prides itself on standing up for language diversity through the use of the 24 official languages in the main EU institutions. From a practical point of view, this position involves a major challenge that deserves closer attention.

For instance, in the European Parliament, parliamentary documents are published in all the official languages “as EU citizens must be able to read legislation affecting them in the language of their own country” (European Parliament 2020) and members of the European Parliament have the right to speak and write in any of the official languages. Rule 167 of the Rules of Procedure of the European Parliament is related to languages, and specifies that: (i) all documents of Parliament shall be drawn up in the official languages; (ii) all members shall have the right to speak in Parliament in the official language of their choice; (iii) interpretation services shall be provided and (iv) the President of the Parliament shall rule on any alleged discrepancies between the different language versions (European Parliament 2021).

As for the citizens of the EU, according to the Treaty on the Functioning of the European Union (TFEU), all European citizens have the right to address the official EU institutions in any of the EU's official languages and to receive an answer in that language. This is intended to make the EU institutions more democratic and accessible to EU citizens. Other provisions related to multilingualism in the TFEU are contained in articles 20, 24 and 342.

Some people think that 24 official languages is too many, and others that 24 official languages is not enough. Some countries try alternative approaches. For instance, Catalan, Euskara and Galician, all spoken in Spain, are considered “additional languages" by the EU (they are co-official languages together with Spanish in their respective territories). This status means that any communication from an EU citizen in these languages has to be translated in Spain into a “procedural language" of the EU, and the answer from the EU institution will be also translated from the procedural language into the additional language. The cost of these translations is borne by Spain.

The use of three procedural languages, English, French and German, is intended to simplify multilingual communication in the EU: given 24 official languages, the EU is faced with a total of 552 possible translation combinations, "since each language can be translated into 23 others" (European Parliament

"The most recent, consolidated version of the TFEU is available from https://eur-lex.europa.eu/ legal-content/EN/TXT/PDF/?uri- CELEX:12012E/TXT&from-EN. Unless otherwise indicated, all urls mentioned in this chapter were more recently accessed in January 2022.

2020) and this would be difficult to handle for all EU documentation. For this reason, there are norms to establish which documents are translated into the other 23 languages and which are translated just into the three procedural languages.

The European Commission's Directorate-General for Translation (DGT) translates texts for the institutions and the citizens of the EU. As of 2022 it produces more than 2.75 million translated pages per year, 91% of which are translations from English, 2% from French, just under 1% from Spanish, and slightly less again from German. Other source languages combined account for around 5% of translation activity. Of all translated documents, 63% are translated internally by the DGT and 37% are outsourced to external companies. Some 55% of translations involve EU law-making, 22% external communication and the web, 12% communication with other EU institutions and national parliaments, 5% correspondence with EU citizens, 4% other official documents, and 2% public consultation on EU policies. The translation budget for 2022 was 355 million euros, or 0.2% of the whole EU budget (European Commission, Directorate-General for Translation 2022).

2.2 Machine translation enters the fray

The DGT has an in-house staff of 2,000, between linguists and support staff, and works with several thousand selected external translators (European Commission, Directorate-General for Translation 2022). The translations they produce, in all language combinations, are stored in the Euramis system (the EURopean Advanced Multilingual Information System), which includes, for instance, the Acquis Communautaire, a corpus of the EU's legislative documents in all 24 of its official languages (European Commission, EU Science Hub 2022). In order to increase productivity and reduce costs, the DGT has incorporated machine translation into some of its workflows, most recently using a system called eTranslation.? The use of eTranslation is expected to save time and money for the EU, but not only that. Eventually, as machine translation develops further, it could contribute to an increase not only in the number of documents that are translated and that otherwise would not have been considered for translation, but also, at some time in the future, to an expansion of the set of languages for which translation is available, and hence to a better reflection of European language diversity; an ideal that, without machine translation, would have been inconceivable just a few years ago. This might be the only way that indigenous or regional minority languages, as well as “non-territorial” or even immigrant minority languages, will gain representation in the EU institutions alongside official, national languages.

?https://ec.europa.eu/info/resources-partners/machine-translation-public-administrationsetranslation en

2.3 What does multilingualism mean to the EU?

Policies on multilingualism are a way of organizing the above-mentioned language diversity, and affirming its richness. In the EU, multilingualism is also seen as a means of social cohesion and worker mobility: “[l]anguage competences contribute to the mobility, employability and personal development of European citizens" (Council of the EU 2014).

A multilingual approach to linguistic policies aims at promoting languages not only in multilingual states but also within organizations. We may talk about multilingualism in many forms, including:

e A multilingual policy, which is the policy of an organization, company or institution to use more than two languages for its internal and external communication.

e A multilingual European Union, which means that different languages coexist in the EU.

e A multilingual citizen, who has the capacity to use several languages.

e A multilingual health system, which is a health system which incorporates linguistic diversity to improve health delivery to newly arrived people, for instance.

The main EU policies on multilingualism are reflected in a series of documents (European Council 2002; European Commission 2008; Council of the EU 2008a,b; 2011; 2014; European Council 2017).

In these EU policies and pronouncements on multilingualism, constant reference is made to language learning and, specifically, the “mother tongue plus two" policy, according to which citizens would learn “at least two foreign languages from a very early age" (European Council 2002). But would implementation of this policy make European citizens multilingual enough? From my point of view, there is a need to incorporate further elements into European policies on multilingualism.

Multilingualism is related to language policies, but that is not the full story. Cenoz (2013), who provides a wide spectrum of definitions of multilingualism, reminds us that multilingualism is multidimensional, involving, for example, the individual versus the social dimension, the proficiency versus use dimension, bilingualism versus multilingualism, etc. It can also be applied to geographical areas or social spheres. Moreover, it can be studied from different perspectives, including those of cognition, social construction, identity, language practices, multimodalities and technologies, among others. From the simplest definition in Wikipedia (“the use of more than one language, either by an individual speaker or by a group of speakers") to the more complex multidimensional definitions provided by Cenoz (2013), in this chapter I will adopt the European Commission's definition of multilingualism as "the ability of societies, institutions, groups and individuals to engage, on a regular basis, with more than one language in their day-to-day lives" (European Commission 2007). The term engage allows us to incorporate a useful nuance for the purposes of this book. By writing this chapter, I wish to invite readers to consider whether multilingualism can be defined from a technological perspective, adapting the above-mentioned definition used by the European Commission.

Discussion topic

Is there a "technological multilingualism", understood as the ability of societies, institutions, groups and individuals to engage, on a regular basis, with more than one language in their day-to-day lives, through multilingual translation tools?

Interestingly, the above-mentioned report from the High Level Group on Multilingualism, which provided our definition of multilingualism, mentions "the potential of multilingual electronic tools as support for non-specialist users of second and third languages" (European Commission 2007) as a research area. Likewise, the European Commission's communication on “Multilingualism - an asset and a commitment" (European Commission 2014) claims that "the language gap in the EU can be narrowed through the media, new technologies and translation services". This book aims to make a contribution precisely to this field.

2.4 EU actions for linguistic diversity

The EU's webpage on linguistic diversity? mentions the following initiatives to promote linguistic diversity: the European Day of Languages, Erasmus+ Mobility programmes, the European Capitals of Culture and the Creative Europe programme:

?https://ec.europa.eu/education/policies/linguistic-diversity en

Established by the Council of Europe in 2001, European Day of Languages‘ takes place on September 26^ each year. On this day EU countries organize activities to promote linguistic diversity and the ability to speak other languages.

e Erasmus + Mobility programmes. Between 2014 and 2020, €14.7 billion was assigned to more than 4 million mobility grants, 2 million of which were designated for university students.

* The European Capitals of Culture is an initiative to highlight the diversity of cultures in Europe, including linguistic diversity. For instance, in 2020, European Capitals of Culture (and corresponding languages) were Rijeka (Croatian) and Galway (Irish and English).

Creative Europe.? This European Commission framework programme supports the culture and audiovisual sectors, including literary translation. Specifically, it funds the translation of literary work from one European language to another.

e Another interesting initiative to promote linguistic diversity is the “European Language Label" attached to EU funded projects. Although most of these projects are oriented towards language learning, some specifically target language diversity.

Discussion topic

From the current perspective, machine translation is not present enough in the EU discourse of language diversity and multilingualism. How could machine translation be included in language learning projects?

Including machine translation in multilingualism initiatives would allow us to increase the number of languages European citizens could become familiar with. It would also allow citizens to approach unknown languages with curiosity and without fear, to access unfamiliar language environments more easily and to respect local languages. Moreover, MT could be used to support reading comprehension in any unlearned languages.

^https://edl.ecml.at ?http://www.creativeeuropeuk.eu/funding-opportunities/literary-translation-0 ®https://ec.europa.eu/education/initiatives/label/label_public/index.cfm

2.5 Multilingualism and language learning in the EU

According to the Council of Europe's Language Policy Portal:

language learners/users lie at the heart of the work of the Language Policy Programme. Whatever their status, all languages are covered: foreign languages, major languages of schooling, languages spoken in the family and minority or regional languages, as well as a specific programme on the linguistic integration of migrants and refugees.

Initiatives to foster multilingualism are many and varied, but language learning deserves closer attention, especially given the EU's above-mentioned “mother tongue plus two" policy.

Some of the EU's recent initiatives to improve language skills include the European Centre for Modern Languages (www.ecml.at; the Eurydice Report (Eurydice 2019), which provides information on policy efforts in Europe that support the teaching of regional or minority languages in schools; the Online Linguistic Support (OLS) platform;? the Common European Framework of Reference for Languages (CEFR); and Erasmus+ mobility programmes.

European projects funded to improve language learning deserve special attention. Methodologies, languages and countries involved vary enormously from one project to another. Table 2 lists some interesting examples.

Eurostat, the website for European Statistics,? provides statistics on the second and foreign languages studied by pupils at different education levels in the EU. According to Eurostat data from 2019, English was by far the most popular language at lower secondary level, studied by nearly 86.8% of pupils, followed by French (19.4%), German (18.3%) and Spanish (17.5%) (Eurostat 2022).

Discussion topic

In second language learning in Europe, is “multilingual” a euphemism for "English-speaking"?

Another interesting question is how many students learn two or more foreign languages, as recommended by the European Council (2002): it is known that

"https://www.coe.int/en/web/language-policy/home *https://erasmusplusols.eu/en/about-ols/ ?https://appsso.eurostat.ec.europa.eu

Table 2: Examples of European projects focused on language learning

Project iTongue: Our Multilingual Future (2013)

Massive open online courses with videos for palliative clinical field and intercultural and multilingual medical communication (2014)

Crafting Employability Strategies for HE Students of Languages in Europe (2015).

LMOCCS for university students on the move (2018)

Gamifiying Academic English Skills in Higher Education: Reading Academic English App (2016).

E-LENGUA: E-Learning Novelties towards the Goal of a Universal Acquisition of Foreign and Second Languages (2015).

EULALIA: Enhancing University Language courses with an App powered by game-based Learning and tangible user Interfaces Activities (2019).

Languages

Not specified

DT, EN, FR, IT, ES, RO

Not specified

FR, ES

Not specified

EN, AR, ES, FR, DE, IT, PT

IT, PO, ES, MT

Specific features

Paralinguistic digital tools for foreign language learning

Embedding employability within language teaching

Open educational resources for university students

Game-based application to improve English academic reading skills of university students

Best practices of the integration of digital competences into the teaching of languages

Inclusive learning tools based on the paradigm of Mobile Learning and Game-Based Learning methodology and the application of Tangible User Interfaces (TUIs)

89.9% (almost 14 million) of secondary level pupils studied more than one foreign language in 2019 (Eurostat 2022). Among them, more than 7 million (48.1%) studied two or more foreign languages.

In short, of the 600 languages spoken around Europe by more than 700 million speakers (EU and non-EU) (World Bank 2020), the majority of EU students are learning one or two out of the following four as their first or subsequent foreign language: English, French, German or Spanish.

2.6 Levels of attainment

The European Survey on Language Competences (European Commission 2019) tested 54,000 secondary pupils (aged 14-15) in 16 educational systems, and covered the two most widely taught foreign languages in all concerned education systems. The survey tested writing, reading and listening comprehension. It did not test oral expression. The key finding of the survey was that only 42% of tested students reached the level of independent user (B1 and B2 in the Common European Framework of Reference for Languages) in their first foreign language, and only 25% reached this level in their second foreign language. Moreover, a large number of pupils did not even achieve the level of a basic user: 14% failed to achieve this level for their first foreign language and 20% failed to achieve it for their second foreign language (European Commission 2019).

In the same report, there are data from a 2018 flash Eurobarometer survey among 15-30 year-olds, where 85% of respondents stated that they wished to improve their proficiency in a language they had already learned (mainly English):

This indicates that the survey respondents were not satisfied with the level they achieved at the end of compulsory education or they did not have a chance to maintain their level. One third of surveyed young Europeans said they were unable to study in a language other than the one they used in school (i.e. often the mother tongue). (European Commission 2019: 102)

2.7 Is there a role for machine translation in language learning?

We have already seen that language-learning efforts in the EU tend to be concentrated on a small number of large languages, and that learners do not always reach desired levels of competence in their chosen foreign languages. These circumstances suggest that further support for language learning is needed, and it behoves us to investigate whether such support could come in the form of machine translation. As neural machine translation learns faster than any foreign language learner, it could, in theory, be used to help learners read complex texts and develop more advanced written skills in their second language. They could learn how to make the most of machine translation in the second language so that they could detect and edit machine translation mistakes based on their knowledge of the second language. And while empirical studies of the use of machine translation in language classes are still thin on the ground, a small number of sources suggest interesting avenues of research. Relevant studies are discussed in Carré et al. (2022 [this volume]), which includes further ideas and strategies that can be used in language learning classes. Yet others are included in the database of activities of the MultiTraiNMT project (Multi TraiNMT 2020). My view is that there are many ways of using machine translation in language learning classes and there is no need to forbid its use if it is used in a conscious and critical way.

On occasion, however, there is just no time to train second language students. Indeed, in the history of machine translation there have been many occasions on which research was partially triggered by a perceived lack of people learning a particular foreign language. Cold War research into Russian-English machine translation is one such case (Gordin 2016). More recently, the Japanese organizers of the Tokyo 2020 Olympic Games (actually held in 2021 due to COVID-19) realized that learning Japanese was out of the question for most foreigners and that they needed a faster approach to overcome language barriers during the Olympics. The Japanese internal affairs ministry thus allocated Y1.38 billion to machine translation research to improve the quality of real-time speech translation technology, with the aim of covering 90% of the language needs of the Olympic teams and tourists who, it was hoped, would go to Japan (Murai 2015). The Japanese government funded the research for a specific machine translation system to be used during Tokyo Olympics and private companies were tasked with the development of devices and mobile apps to run the system. The plan was that companies would recover their investment by selling the devices and apps subscriptions to users. In this case, the introduction of machine translation was the chosen shortcut to bring multilingualism to Japan, instead of language learning.