Machine translation (MT) is software that renders text from one language into another automatically. For language teaching the question has shifted from whether learners use it to how. Tools such as Google Translate and DeepL run on neural networks that translate whole sequences rather than word by word, and the jump in fluency this brought has moved MT from a banned crutch toward a tool that, used deliberately, can support learning rather than short-circuit it.
The older reflex was prohibition. MT looked like cheating: paste the prompt, copy the output, submit work the learner could not have produced. Neural MT made the output good enough that a blanket ban became both unrealistic and wasteful, since learners use the tools anyway and the more interesting question is what they can be taught to do with them. This is not a revival of the grammar-translation method, where translation was the syllabus and accuracy in rendering sentences the goal. Here translation is a means: the target stays communicative competence, and MT is one resource learners learn to use critically, much as digital literacy reframes any powerful tool around informed judgement.
Several uses have a defensible rationale. As comprehension support, MT lets a learner get the gist of a text pitched above their level and stay in the target language longer. As a vocabulary look-up it is faster than a bilingual dictionary, though it hides the polysemy a dictionary entry would expose. More ambitiously, MT can drive noticing and contrastive work: comparing a learner's own draft with an MT rendering surfaces where the two languages diverge, building the kind of cross-linguistic awareness that explicit attention to form depends on. Post-editing tasks, where learners revise raw MT output into accurate, appropriate text, turn the tool's errors into the material of the lesson and demand exactly the judgement the learner is meant to build.
Research documents heavy, largely unsupervised use. Sangmin-Michelle Lee's study of Korean EFL writers found that MT reduced lexical and grammatical errors and lowered the burden of producing text, while also encouraging some learners toward a shallow engagement with structure when used without guidance. Jason Jolley and Luciane Maimone's review of three decades of MT in language education traces the field from early scepticism to cautious, conditional acceptance and maps where the evidence is thin. Emily Hellmich and Kimberly Vinall's survey of US foreign-language instructors found teachers drawing firm but inconsistent lines around acceptable use (by text length, by skill, by stated policy) while acknowledging that student use is widespread and driven by varied motives, and that MT is felt as both resource and threat to the profession.
The first risk is accuracy paired with over-trust. Neural output is fluent enough to look right, which makes its errors harder to catch precisely for the learners least able to spot them; idiom, register, and culturally loaded meaning are where it slips. Dependency is the second: a learner who routes every difficult sentence through MT may never build the retrieval and problem-solving that production requires, a deskilling that mirrors the autonomy concerns running through AI more broadly. Academic-integrity questions are unavoidable when MT output is indistinguishable from independent work, and instructors are poor at detecting it. Equity cuts both ways: free tools widen access for under-resourced learners, yet the best engines and the literacy to use them are unevenly distributed. The constructive response, argued by Lynne Bowker, is MT literacy: teaching learners to decide whether, when, and why to use MT, to prepare text so the engine handles it well, and to post-edit the result critically, so the tool becomes an object of instruction rather than a hidden workaround.
Set a transparent policy and teach to it rather than policing in the dark; learners use MT regardless, so the gain is in shaping how. Build tasks that make the tool visible — post-editing, MT-versus-own-draft comparison, error hunts in raw output — so its limits become the lesson. Use it to extend comprehension and keep learners in the target language, but pair vocabulary look-ups with attention to the senses MT flattens. Treat principled first-language use and MT as continuous: both are bridges, valuable when they feed noticing and costly when they replace it. Above all, make MT literacy an explicit objective, since the durable skill is critical use, not avoidance.