Is Machine Translation a Reliable Tool for Reading German Scientific Databases and Research Articles?

Area of Science:

• Chemistry
• Scientific Information Management

Background:

• Numerous scientific databases and research papers remain untranslated, particularly in German.
• Key German resources like Beilstein and Gmelin contain historical chemical information not always duplicated in modern databases.
• Accessing these resources requires professional translation or German language proficiency.

Purpose of the Study:

• To evaluate the reliability of machine translation (MT) for scientific German to English translation.
• To compare the performance of different neural network machine translation (NMT) platforms for scientific texts.

Main Methods:

• Selected German scientific texts from various historical periods and sources.
• Tested Google Translate and DeepL for German to English translation capabilities.
• Compared machine translations against human translations for accuracy, especially for complex sentences and synthetic procedures.

Main Results:

• DeepL Translator and Google Translate demonstrated high reliability in translating diverse technical passages from German scientific literature.
• Both platforms proved effective for a wide range of scientific texts spanning over a century.
• Identified potential inaccuracies in MT for specific details, such as reagent addition in synthetic methods.

Conclusions:

• Neural network machine translation tools like DeepL and Google Translate are valuable for accessing historical German scientific literature.
• While generally reliable, caution is advised when using MT for critical experimental details, particularly in organic synthesis.
• Further careful review is recommended for machine-translated scientific methods and procedures.