With the increasing number of research articles by non-native English speakers and a lack of spare time in researchers' schedules, automated translation may seem like an appealing option. Similar to automated proofreading, this type of system is readily available and often low cost or free. Google Translate, which bases its rapid translations on patterns of word usage in other published and translated documents, is one commonly used tool. However, machine translation frequently presents several types of errors that can affect a reader’s understanding. The four arguably most serious mistakes are as follows:
- Sentence fragments
- Long sentences
- Illogical ordering of phrases
- Literal instead of context-dependent translation
Poor sentence construction, syntax, and terminology use may decrease the readability of a text, leading to unclear or even lost meaning. These issues increase the likelihood of subsequent rejection by professional language editors and journal editors alike.
To demonstrate these flaws of automated translation, part of the abstract of a scientific paper (Yves et al., 2010; CC-BY license) written in French was translated into English using Google Translate. This machine translation was then compared with the English-language translation provided by the paper’s authors:
“Introduction : Malgré l’essor des techniques chromatographiques couplées à la spectrométrie de masse, l’immunoanalyse revêt toujours une importance capitale dans les laboratoires hospitaliers et de médecine légale. Nous nous attacherons ici à décrire quels pourraient être les protocoles à mettre en œuvre pour utiliser ces tests immunochimiques sur des matrices non conventionnelles (sang total laqué, cheveux…) et sur les préparations nécessaires pour analyser ces échantillons. Différentes techniques immunochimiques avec leur aptitude à accepter des matrices dites alternatives seront évoquées.”
“Introduction: Despite the development of chromatographic techniques coupled with mass spectrometry , immunoassays always of paramount importance in hospital laboratories and forensic medicine. Here we will describe what might be the protocols to implement these immunochemical tests for use on non- conventional matrices (lacquered whole blood, hair ... ) and the necessary preparations to analyze these samples . Methods: Different techniques immunochemical with their ability to accept so-called alternative matrices will be discussed .”
“Introduction: Despite the breakthrough of chromatographic techniques coupled to mass spectrometry, immunoanalysis is still a key component of the various techniques used in clinical or forensic toxicology. This article describes different procedures to use immunoassays on unconventional biological matrices such as clotted blood or hair specimens. A focus is made on sample preparation needed to analyse such matrices. Various immunoassay techniques and their ability to accept unconventional matrices are discussed.”
Although the published translation appears to have taken certain stylistic liberties with the original text (such as by translating “les laboratoires hospitaliers et de médecine légale” as “clinical or forensic toxicology” instead of “hospital laboratories and forensic medicine”), these changes do not seem to have significantly altered the authors’ intended meaning. In contrast, the machine translation changes both the clarity and the meaning of the text for the four reasons outlined above.
1. Sentence fragments. The machine translation of the second clause of the first sentence omits the verb “are” (“immunoassays always of paramount importance in hospital laboratories and forensic medicine”), decreasing the fluidity of the prose.
2. Long sentences. Considering the second sentence, the phrasing in the machine translation is somewhat clumsy: “Here we will describe what might be the protocols to implement these immunochemical tests for use on non-conventional matrices...and the necessary preparations to analyze these samples.” In this case, the translation is not inaccurate; rather, it is insensitive to the English-language custom of concise writing (whereas French tends to favor longer and more elaborate sentences). In contrast, the human translator broke this sentence into two and decreased its wordiness, thereby enhancing clarity.
3. Illogical ordering of phrases. The phrase “Different techniques immunochemical” in the machine translation clearly should be “Different immunochemical techniques.” The difference in word order conventions between French and English would be familiar to a human translator.
4. Literal instead of context-dependent translation. This problem particularly arises when a commonly used word has a different meaning in a technical context. In the example above, the French term “sang total laqué” is translated as “lacquered whole blood” by Google Translate, even though “laqué” actually means “clotted,” and not “lacquered,” in this field-specific context. To circumvent this type of misunderstanding, a human translator may gauge meaning by assessing the immediately surrounding text, by further reading the paper, or by referring to conventions in other papers in the same field.
Further examination reveals a few additional, more minor issues with automated translation: punctuation errors (such as the extra spacing around the ellipsis) and insensitivity to the authors’ preference for American or British English (in this particular case, “analyse” was stated in the published translation, but the machine used “analyze”).
Therefore, to avoid obscuring or altering your meaning when translating your manuscript, consider seeking only human translation help. During the process, you can refer to English-language papers in your field for specific terminology, and a thesaurus may also be helpful in determining the right word choice. Additionally, collaborating with a colleague who is a native or near-native English speaker may be useful for achieving a fluid translation. Alternatively, a specialized professional translator with expertise in your research area may be able to most efficiently and accurately translate your work.
Read part one of this series, Automated Proofreaders: Human vs. Machine.