Article type
Year
Abstract
Background: Volunteers from the Brazilian Cochrane Center (BCC) translate English Cochrane abstracts into Portuguese, a language and a content expert revise the texts and these official translations are then uploaded to the Cochrane Library. However, users can also find unofficial translations of Cochrane abstracts on the web, which are produced by others using translation software.
Objectives: To appraise critically the quality of one abstract translated by automated software and available on the web and compare it to the translation produced by the BCC team.
Table. Number of sentences correctly translated by machine and by humans
Methods: We compared the original English abstract (CD007438) with two machine translated versions ('E-version 1 and E-version 2') available at www.epistemonikos.org. We then compared these two versions to the version translated and revised by the BCC team ('BCC version').
Results: The title of E-version 1 had word inversions and was incomprehensible. Of the 18 sentences in the original abstract, E-version 1 had only seven correct sentences, while in 11 there were word inversions, which hindered the comprehension and changed the meaning completely. E-version 2 also had sentences with changes in meaning (Table 1). In the BCC version, the meaning and comprehensibility of the text were correct, the number of errors was smaller and most of the corrections were related to style.
Conclusions: While machine translations can reduce the workload involved in Cochrane translations, the reviewer of these texts must be experienced with the Cochrane vocabulary and fluent in the language of translation, to ensure that the final version of the text is comprehensible to native speakers and conveys the exact information provided in the original English version. The use of rigorous methods in the revision of machine or human translations is the only way to ensure that users have access to high quality information.
Objectives: To appraise critically the quality of one abstract translated by automated software and available on the web and compare it to the translation produced by the BCC team.
Table. Number of sentences correctly translated by machine and by humans
Methods: We compared the original English abstract (CD007438) with two machine translated versions ('E-version 1 and E-version 2') available at www.epistemonikos.org. We then compared these two versions to the version translated and revised by the BCC team ('BCC version').
Results: The title of E-version 1 had word inversions and was incomprehensible. Of the 18 sentences in the original abstract, E-version 1 had only seven correct sentences, while in 11 there were word inversions, which hindered the comprehension and changed the meaning completely. E-version 2 also had sentences with changes in meaning (Table 1). In the BCC version, the meaning and comprehensibility of the text were correct, the number of errors was smaller and most of the corrections were related to style.
Conclusions: While machine translations can reduce the workload involved in Cochrane translations, the reviewer of these texts must be experienced with the Cochrane vocabulary and fluent in the language of translation, to ensure that the final version of the text is comprehensible to native speakers and conveys the exact information provided in the original English version. The use of rigorous methods in the revision of machine or human translations is the only way to ensure that users have access to high quality information.