03 · What You Need to Know
A Language Barrier Is a Research Problem to Manage, Not a Reason to Pretend the Source Does Not Exist
First, determine how much of the source you actually need
Not every encounter with a foreign-language source requires a professional translation of the entire document.
Perhaps you need only to determine whether the study is relevant enough to retrieve. You may need to understand its research question and principal findings. Alternatively, the paper may be central to your evidence synthesis, requiring accurate extraction of its methods, sample characteristics, outcome definitions, statistical results, limitations, and conclusions.
These are different translation tasks with different consequences if something goes wrong.
| What you need from the source |
Possible starting approach |
Level of caution |
| Determine whether the paper is relevant |
Translated title, abstract, keywords, or preliminary machine translation |
Moderate |
| Understand the broad argument or findings |
Machine translation of relevant sections, preferably checked against available English metadata or summaries |
Moderate to high |
| Extract exact numerical results |
Full translation with careful verification of tables, units, statistical terminology, and surrounding text |
High |
| Evaluate methodological quality |
Detailed translation of methods and results with fluent or specialist verification when consequential |
High |
| Quote exact wording |
Human-verified translation and transparent treatment of the translation |
High |
The useful principle is proportionality. A translation used only to decide whether an article deserves closer examination does not need to carry the same evidential burden as a translation from which you extract an effect estimate for a meta-analysis.
Check whether another language version already exists
Before translating the paper yourself, determine whether someone has already solved the problem.
The journal may provide the abstract in English even when the full article is in another language. Some journals publish bilingual articles or provide translated versions. An author may have deposited a manuscript in another language. A thesis or report may have an official translated edition.
Search the article title, DOI, authors, and journal. Check the publisher page and supplementary files. Search Google Scholar for other versions. If the paper is available through a repository, inspect its metadata for alternate files or language information.
When you can find only the citation or one inaccessible manifestation of the source, the broader strategies for locating the full text from a bibliographic record may reveal another usable version.
Do not assume that the English abstract represents the entire paper
An English-language abstract can be extremely useful. It may tell you enough to establish the study's topic, design, population, and principal reported findings.
But an abstract remains a compressed representation of the full study regardless of language. It may omit detailed methods, secondary outcomes, qualifications, limitations, or contextual information needed for your particular claim.
If the abstract is all you can read, the same evidential limits apply as when citing any paper after reading only its abstract. Do not attribute information to the full study that you have not actually verified.
Machine translation can be useful, but its reliability is task-dependent
Machine translation has become a practical tool for researchers who encounter literature outside the languages they read. Evidence suggests that it can sometimes provide sufficiently useful translations for scholarly tasks, but performance is neither uniform nor error-free.
A study of Google Translate for systematic-review data extraction translated randomized trials from Chinese, French, German, Japanese, and Spanish into English. Researchers were able to extract substantial amounts of study information, but accuracy varied by language and type of information. Extraction was generally more reliable for study-design and intervention information than for outcome descriptions and results. The authors concluded that machine translation could help reduce language bias but warranted caution because translation errors remained possible.
A later study comparing DeepL, Google Translate, and CUBBITT on French medical research abstracts found broadly similar performance across the systems on the automated metrics used, although the authors emphasized that their evaluation involved abstracts rather than complete specialized articles.
These findings support a nuanced position: machine translation can be genuinely useful for research, but "the software produced fluent English" is not evidence that every technical detail was translated correctly.
Fluent translation can still be wrong
This is one of the most important practical cautions.
A translated passage can read naturally while containing a semantic error. The danger is greater when you cannot compare the translation with the source language yourself because fluency may create unwarranted confidence.
Technical terminology, negation, numerical expressions, abbreviations, discipline-specific meanings, and syntactically complex sentences can be particularly consequential. Research on machine translation of scientific material has documented cases in which apparently plausible translations altered technical meaning.
Watch Out
Do not use grammatical fluency as your accuracy test. A machine-translated sentence can sound entirely natural while mistranslating a technical term, reversing a relationship, mishandling negation, or obscuring the conditions under which a result applies.
Numbers are not automatically safe just because they do not need translation
A value such as 0.42 looks the same in many languages, but understanding what that number represents requires language.
Is it a correlation coefficient, odds ratio, standardized effect, probability, prevalence estimate, regression coefficient, or something else? Does the table report percentages or proportions? Is the value adjusted or unadjusted? Does a footnote restrict the sample? Does a minus sign or decimal convention have a particular meaning?
The systematic-review research on machine translation is relevant here because extraction errors were especially concerning for results information. Numerical transcription without accurate interpretation of labels and surrounding text can create a false sense of precision.
Translate tables together with their headings and notes
If you need data from a table, do not translate only the row containing the number you want.
Translate the table title, column headings, row labels, abbreviations, footnotes, statistical notes, units, and any nearby text explaining the analysis. A result separated from those elements can easily be misclassified.
The same applies to figures. Axis labels, legends, captions, subgroup definitions, confidence intervals, and symbols may contain information necessary to interpret what the graphic actually shows.
Methods deserve more verification than general background prose
A minor translation error in a literature-review sentence may be inconvenient. A translation error in a dosage, inclusion criterion, intervention protocol, measurement procedure, or statistical method can change your understanding of the study.
If your research depends on methodological details, translate the relevant methods carefully and seek human verification when uncertainty remains.
This is particularly important in systematic reviews, clinical research, historical interpretation, legal or policy analysis, and other settings where a small linguistic difference can materially alter classification or inference.
Use a bilingual colleague strategically rather than asking them to translate everything
If you have access to a colleague, librarian, research assistant, or collaborator who reads the language, you may not need to ask for a full translation.
Instead, identify the passages where uncertainty matters most:
- the eligibility criteria;
- the exact outcome definition;
- a sentence containing an important qualification;
- a table heading or footnote;
- a technical term translated inconsistently;
- a conclusion that appears stronger than the results;
- the wording of a quotation you intend to reproduce.
This targeted approach can make human verification more feasible while preserving rigor where it matters.
For high-stakes use, consider a qualified translator or domain expert
When the precise meaning of a source materially affects a consequential conclusion, professional translation or verification by someone fluent in both the language and the subject area may be warranted.
Language proficiency alone does not guarantee familiarity with specialized terminology. A fluent speaker who does not know the discipline may struggle with technical concepts, just as a subject specialist with limited language proficiency may miss linguistic nuance.
The ideal level of expertise depends on the task. Translating a general introduction is different from translating a psychiatric diagnostic instrument, a historical legal text, an engineering procedure, or a statistical appendix.
Machine translation can be a first pass rather than the final authority
A practical workflow is to use machine translation to establish the structure and broad meaning of the article, then identify passages requiring higher confidence.
You might first translate the abstract, introduction, methods headings, results, and discussion. If the article proves irrelevant, you can stop. If it becomes important, translate the relevant sections more carefully and verify critical passages.
This prevents you from investing heavily in translating an article that ultimately contributes little to your research while still allowing important non-English evidence to enter consideration.
Compare more than one translation when wording looks consequential
If a machine-translated passage seems ambiguous, technically odd, or unexpectedly important, comparing outputs from more than one translation system can reveal instability.
Agreement between two systems does not prove correctness, and disagreement does not tell you automatically which translation is right. It does, however, flag passages that deserve closer scrutiny.
For important terminology, search how the term is used in bilingual scholarly sources, dictionaries, glossaries, reporting guidelines, or other authoritative disciplinary materials.
Keep the original text available during your analysis
Do not replace the source-language document with a translated copy and then lose track of the original.
Preserve the original file, record which translation tool or translator you used when relevant, and keep enough information to return to the original passage if a question arises later.
This becomes particularly useful during peer review. If a reviewer challenges your interpretation of a translated study, you should be able to identify the exact source-language passage from which the interpretation came.
Do not cite the translation software as though it authored the research
The scholarly source remains the work written by its authors. Translation is the means through which you accessed its content.
How a translated quotation or translated work should be cited depends on the citation style and whether you used a published translation, created your own translation, or relied on another identifiable translator. Citation styles differ, so consult the specific rules required for your manuscript when translated wording appears directly in your text.
If you are merely using machine translation privately to understand a source, the translation tool does not replace the original work as the source of the research claim.
Be transparent when your own translation matters to the reader
If you directly quote a passage and provide an English translation that you created, readers may need to know that the wording is your translation rather than an officially published English version.
Exactly how this should be indicated varies by citation style and discipline. Common approaches include a parenthetical indication such as "author's translation" or an explanatory note, but you should follow the relevant style guide rather than assuming one format applies universally.
For paraphrases based on translated material, the need for explicit disclosure depends on context. A methodological paper describing multilingual evidence extraction may require detailed reporting of translation procedures, whereas an ordinary literature review may not need to document every private translation step.
Language restrictions can affect the evidence base
The problem is larger than one inconvenient article.
If researchers systematically exclude studies because they are not published in languages the research team can read, the resulting evidence base may become linguistically selective. In systematic reviews, this is often discussed as language restriction or language bias.
Machine translation has therefore attracted attention partly because it may allow reviewers to consider studies that would otherwise be excluded for practical linguistic reasons. Research comparing translated and original-language trial reports suggests that this approach can increase accessibility, although the risk of extraction error must be managed.
For a formal review, decisions about language eligibility should be established in the protocol where appropriate and reported transparently. Translation strategy becomes part of the review method rather than an improvised decision made after an inconvenient paper appears.
A language barrier does not automatically justify secondary citation
Suppose an English-language review summarizes a Japanese paper that you cannot read. It may seem easier simply to cite the Japanese study through the review.
That may sometimes be defensible, but language alone should not automatically end your attempt to consult the original. A translated abstract, machine translation, bilingual colleague, translated version, or professional assistance may allow you to examine the source directly.
If the original genuinely cannot be interpreted with sufficient confidence, then secondary citation may be appropriate when the information is necessary and the citation style permits it.
The key is transparency. If your substantive knowledge of the study comes from the English-language review rather than your examination of the Japanese paper, do not make the citation suggest otherwise.
Sometimes the responsible decision is not to use the source for a particular claim
An important-looking paper does not have to be forced into your manuscript at any cost.
If you cannot establish the meaning of a critical passage with sufficient confidence, you can seek independent evidence for the same point, qualify the claim, or omit it.
This is particularly sensible when a single uncertain translation would carry substantial argumentative weight. Research rigor sometimes consists of knowing when the available evidence does not support the sentence you wanted to write.
