When I worked as a food runner in a Washington, D.C., restaurant, I began learning Spanish so I could speak with some of my coworkers in their native tongue. One night, as we were cleaning up, I said to one of my colleagues, Necessito usar la mopa (“I need to use the mop”), which first provoked confusion and laughter, then an ill-fated defense on my part to explain that what I said was correct. And while it may have been grammatically correct, I had failed to make myself understood.
The trouble with translation is that there are no perfect translations, but lots of wrong ones. The trick is to find the best possible approximation that retains both the structure and the meaning of the original phrase.
In beginner language classes, forming simple sentences like this can feel like solving a math problem. You’re wrestling with these different components in your head and trying to manipulate them to produce the best string of articles, nouns, pronouns, verbs and adjectives possible.
When I was in the third grade, circa 2008, my classmates and I started to learn about fractions and division. And as a member of our technological society, I was not about to be tricked into using my mind when my shiny blue TI-15 Explorer™ calculator could do the job.
But what does the calculator as a mere technical object have to do with translation?
The analogy here is between the function of calculators in mathematics and translation software in language learning. The disanalogy is between the skill of calculation and the capacity to understand, interpret and actively engage with the products of language.
When it comes to language, finely-tuned intuition matters more than mere calculation. How ideas are expressed through grammar, syntax and vocabulary is complicated by the fact that language is always tied to social and cultural contexts, as well as what the speaker wants to say within the confines of the target language.
I bombed my first division and fractions test in third grade; it had not occurred to me that the assessments would be a non-calculator affair. Left to my own devices — which, in this case, consisted of a pencil — I felt helpless.
It was also around this time that I started taking Sunday French classes at the Alliance Française in DuPont Circle. Back then, Google Translate was only two years old and still pretty useless. Without a translator, I struggled through a lot of conjugations and awkward conversations before I started forming sentences on my own.
Now, as a French and francophone studies major, I have gained an appreciation of the power of translation tools, as well as their limitations.
Part of this appreciation has stemmed from improvements in software. Google Translate, for example, started out using statistical machine translation that systematically compared documents in different languages from online databases. In 2016, Google revolutionized its translation platform with the announcement of the Google Neural Machine Translation system.
Barak Turnovsky, director of product at Google AI, claimed in 2016 that the particular advantage of Neural Machine Translation is the analysis of entire sentences rather than just phrases or words in its machine learning algorithms. Further, the relatively autonomous “end-to-end learning system learns over time to create better, more natural translation,” Turnovsky wrote at the time.
For many languages, translation tools have made huge advances from when I started learning French. They provide an excellent service for increasing understanding between people of different cultures, but for students, their usage comes with some intangible costs.
To get the student perspective on translation tools, I spoke over the phone with Sara Stawarz, a freshman in the LSA Residential College. Stawarz studies Spanish in the RC, and like many other RC students, is working towards being bilingual.
“That’s my goal,” Stawarz told me during our conversation. “Especially because my family is from Puerto Rico. So I’d love to be able to speak Spanish with my grandma and my extended family.”
Stawarz, like myself, is currently following the RC intensive language program, a semi-immersive experience that includes two semesters of intensive language study (eight credits each), along with one four-credit literature seminar that aims to apply students’ knowledge of the language to a broad range of topics.
Before enrolling in the RC French program, I did not realize I would be speaking, writing and thinking in French for up to three hours a day. My cohort and I also participated in French-only lunches and coffee hours. Thankfully, unlike the LSA language requirement, all Intensive courses were pass/fail; because of this, we felt free to take risks that we might not have taken if the courses had been graded.
My conversation with Stawarz brought back memories of that time spent in East Quad, where one could reliably hear impressions of Spanish, French, German, Russian and Japanese throughout the halls. The pandemic has changed the setting, though the spirit language learning has persisted. When I asked Stawarz if her Spanish has improved, she gave an impressive affirmation.
“I have learned more in my first semester of Spanish here than I did in the entirety of high school Spanish, which is absolutely insane to me,” Stawarz said.
But when I mentioned the use of translation tools in the program, it was clear that they presented a particular ethical and educational problem for her and the other students in her cohort, particularly with the technology an internet tab with virtual learning.
“Say you’re in lecture, and your teacher asks you a question, and you panic,” Stawarz explained. “It’s so easy to just immediately turn to SpanishDict to look up the conjugation of a word instead of internalizing it and trying to think through it yourself. And it leads to this sort of dependency.”
Online translation tools present a constant temptation for students. To be fair, a major part of acquisition is imitation. But language students and professors admit the dangers of developing a dependence on language tools that won’t be there when you need to express something on the fly.
For example, one day in Spanish class, after arriving at Detroit Metropolitan Airport at around 5 a.m. that morning, Stawarz explained to her professor that she took un robo de ojos rojos (“a red-eye flight”). Most Americans would recognize this as an overnight flight, such as those from Los Angeles to New York. Google Translate renders it as un vuelo nocturno (“a night flight”) while DeepL agrees with Stawarz’s literal interpretation, un vuelo de ojos rojos.
In fact, the red-eye flight is not a common concept among hispanophones, though Stawarz said her teacher applauded the effort.
Beginner and intermediate students need to make such efforts in order to progress. For advanced students who have a mastery of the language, however, translation tools can save valuable time and energy.
The use of Google Translate in language classes should be similar to the use of calculators in mathematics. To earn the calculator, I must demonstrate a working knowledge of the underlying principles, rules and conventions related to the problem. It must be a beneficial tool in the short run and the long run: an enhancement of existing skills, not a patch for incompetence.
With languages, however, one needs to master both the script of the language as well as its use in everyday experience. The human translator retains their value because languages connect to the world. As students of ancient or endangered languages will tell you, languages are partly dynamic and depend just as much on the lived experience of speakers as the static definitions found in any dictionary or language textbook.
Languages are alive, and translating from one to another is a matter of politics and experience more than one of mere calculation.
Amanda Gorman’s poem, “The Hill We Climb,” which she performed at the 2021 presidential inauguration at only 22 years old, was an inspiration to many Americans, but especially to Black women and girls who could relate to the experiences Gorman expresses through her words.
Gorman recited her inaugural poem just 14 days after supporters of Donald Trump breached the U.S. Capitol. She wrote the poem in response to that event. The way she captures the feeling of cautious optimism felt by many Americans as the sun rose the day after spoke to me, even though her experience is different from my own.
Abroad, however, there have been numerous controversies surrounding the ability of white translators to translate “The Hill We Climb.” The controversy does not surround the selection of words or phrases so much as the ability to capture the contingent experience of Gorman as a Black woman spoken-word artist.
One day after accepting the position to translate Gorman’s poem from English into Dutch, the 29-year-old Dutch writer Marieke Lucas Rijneveld, who is white and identifies as non-binary, reversed their decision, after a round of criticism from members of the Dutch media as well as Black Dutch spoken-word poets who were not consulted for the role. Similarly, the Barcelona publishing house Univers canceled an agreement with Catalan translator Víctor Obiols, a white man, because he did not fit the “profile.”
Clearly, there is a politics of translation that involves attention not only to syntax, grammar and meaning but to the historical experiences of marginalized groups and to the interplay of language and experience. Homespun translations of “The Hill We Climb” are available on the internet in many languages — and anyone with an internet connection can copy/paste the English text into Google Translate and read from the output. But here we are dealing with a poetic work of significant weight, something that should not be translated lightly.
Behind the calls to align the experience of the translator with that of the author is to ensure that the translator cares about revealing the multiple meanings within the text to new audiences. That is not to say that a white translator is not able to produce a good translation of a work authored by a Black woman, or that they will not exercise care in doing so. However, not all language knowledge can be learned from the outside, and we should not underestimate the essential role of “embodied knowledge” that can only come from experience and culture, nor the value this knowledge can bring to the translated work.
So, the question is not whether a translation tool like Google Translate can produce a translation as good as that of a human. It is, rather, to what extent the translation machine can simulate the embodied experience of the human translator.
To what extent can a neural network care about the words it has reproduced on the screen; in the words of the philosopher John Haugeland, is Google Translate capable of “giving a damn?”
While I think translation tools do pretty well in translating texts of a basic to moderate complexity, they are no replacement for an embodied, experiential knowledge of the target language.
The involvement of my mind, my face and my hands while speaking languages allows me to express myself and constitute a Weltanschauung (“worldview”) that diverges from my anglophone way of seeing the world. The use of translation tools, on the other hand, stifles this embodied-ness, externalizing the processes that allow for authentic expression.
Giving a damn about language learning opens a path for self-growth and responsible citizenship in an increasingly multicultural world. Translation tools can be a helpful guide on this path, but only if we are clear-eyed about their limitations, as well as their disconnection from the experiences that constitute the life of a language.