Beyond the Language Industry: Helping Others to Develop Machine Translation Literacy Skills
For the first 50 or so years of its existence, machine translation technology was primarily in the hands of researchers and language professionals. However, the early 21st century witnessed a number of significant changes to the MT landscape:
- MT has been released “into the wild” with the introduction of free online MT tools, putting this technology into the hands of users with no background or experience in translation;
- MT tools are exceedingly easy to use, often requiring just a single click, or a simple “copy, paste, click”;
- MT continues to garner an increasing amount of attention from the press, although the popular media does not always present a balanced or nuanced view of MT, and the overly positive “hype” about MT’s capabilities may lead users to apply it uncritically;
- MT output has seen dramatic improvements in quality with neural approaches, meaning that the raw output may be usable for some purposes. In particular, the fluidity of neural machine translation means that the output sounds plausible, even if the content is not accurate, and the lack of noticeable “translationese” (e.g. clunky or unnatural sounding passages) makes it harder to spot errors, particularly for those users who are not language professionals.
In short, MT is easy to access and use, and the quality of MT output has improved. However, users do not inherently adopt a critical mindset when using it. In order to become informed and critical users of MT tools, and of their translated output, users need to develop machine translation literacy.
What is MT literacy?
When describing the MT literacy project, I like to say that “Using machine translation is easy, but using it critically requires some thought.” Language professionals usually receive training both in translation and in how, when, and whether to use translation technology, as well as how to recognize and compensate for its limitations. However, people outside the language professions do not typically receive such training. As noted above, the how-to skills of using machine translation tools are very easy to acquire since they consist of little more than copying and pasting a text, choosing a language pair, and clicking the “Translate” button. However, MT literacy is less about developing techno-procedural skills and more about encouraging critical thinking tasks, such as evaluating the suitability of a text for translation by machine or weighing the benefits and risks of using MT against other translation solutions.
What information could be shared in an MT literacy training session?
MT literacy is a customizable concept, so the nature and volume of information to share depends on the target audience. A group of teens might need different information than people who teach English-as-a-second language at the postsecondary level, and so on. The following list contains ideas for possible elements to include in an MT literacy training session:
- Explain the general concept of machine learning and the overall neural approach to machine translation (i.e., it’s a data-driven approach);
- Consider the type, quantity, and quality of data required for data-driven neural machine translation, identify how machine translation systems can be sensitive to data, and outline the potential consequences of data insufficiency (e.g. high/low resource languages and domains, and algorithmic bias);
- Describe the need for transparency with regard to machine translation use (e.g. academic integrity, fair use, sustainability, labeling of MT output);
- Conduct basic risk assessment regarding machine translation use (e.g. high-stakes tasks such as a doctor’s visit vs low-stakes tasks such as translating song lyrics);
- Compare and evaluate the results produced by a selection of free online machine translation systems (i.e., looking beyond Google Translate);
- Modify input texts to reduce ambiguity and improve the quality of the machine translation output (i.e., translation friendly writing and the principle of “garbage in, garbage out”);
- Apply basic post-editing techniques to improve machine translation output according to fit-for-purpose principles.
Who has been helped by MT literacy instruction?
Beginning in 2019, I started to make a conscious effort to assist university students who were not part of a translator training program to develop MT literacy skills. While many of the participating students are international students who have come from other regions of the world to study in Canada, some Canadian students have also participated, including both English- and French-speaking Canadians as well as those who speak Indigenous or heritage languages. These efforts to deliver MT literacy instruction have taken a number of different forms and branched out to include additional audiences, such as:
- A module on courses for first-year undergraduates aimed at learning information literacy skills
- Short workshops offered in partnership with the university library and the international students office
- Guest lectures for students in English-as-a-second language courses
- A professional development workshop for teachers of English-as-a-second language
- A short course on machine translation for students at a Digital Humanities summer institute
- Courses on translation for non-translators, which go beyond learning about MT, but which nonetheless include an MT literacy component.
For some of these activities, I asked participants to complete a short voluntary and anonymous survey at the end of the MT literacy training session in order to get feedback. Figure 1 shows the results from 170 undergraduate students who participated in a course on translation for non-translators. In response to the question of whether MT literacy is relevant for people outside the language professions, over 70% identified it as being very important or essential.
How can the language industry help others to develop MT literacy?
The efforts of one individual can only go so far, but working together, the members of the language industry can have a much greater impact. Some organizations, such as the American Translators Association, have a formal outreach program where they visit to schools to share their knowledge of translation with students. This type of model could be adapted to include MT literacy, and inspired by this model, as well as by the observations of Andrew Joscelyne on the TAUS blog, I recently participated in an MT literacy outreach activity for teens. In the context of Science Literacy Week, I worked with the University of Ottawa’s Faculty of Engineering Outreach team to offer an online session to de-mystify machine translation for high school students. Other venues for outreach could include the public library, seniors groups, or youth groups (e.g. Girl Scouts).
Another effective way to do outreach could be for language professionals to participate in the delivery of continuing professional development education for groups such as teachers, so that these teachers may in turn offer MT literacy instruction to their students. Such professional development could even be extended to groups other than teachers, such as healthcare workers, journalists, people who work with NGOs that assist immigrants or refugees, patent professionals, or indeed any other profession who may use MT in the course of their work.
Translation or localization companies, professional associations, or MT tools developers, could add some basic MT literacy information or tips to their websites. Working together, members of the global language industry have the capacity to help a lot of different users to develop or improve their MT literacy skills. This will help them to become more informed digital citizens, and that’s good for our society as a whole. Indeed, one of the United Nations’ Sustainable Development Goals is quality education, which seeks to “ensure inclusive and equitable quality education and promote lifelong learning opportunities for all” (SDG4). The language industry has a lot to offer in this regard, so what are we waiting for? Let’s do this!