What to expect in Machine Translations and guidelines for completing editing tasks
Reading and editing a machine translation can be quite different than working with a translation fully completed by humans.
What is a Machine Translation?
Machine translation is the use of computer software to automatically translate text or speech from one language to another. This technology relies on complex algorithms (formulas) and large databases of linguistic information to translate words and sentences accurately.
How are Avodah's machine translations of the Bible produced?
In Phase 1 of a translation project, we collect translation of 1200 verses from the Bible, plus translations of other documents, and sometimes certain books from the Bible. Those 1200 verses have many of the words and ideas that are found throughout the whole Bible. The AI ‘brain’ learns then language from the 1200 verses and other collected data and then, based on that information, makes a translation of the WHOLE Bible.
What types of errors can I expect to see in a Machine Translation (MT)?
A machine translation is never perfect in its first draft. This is why it is important to include the process of Machine Translation Post-Editing (MTPE). MTPE is the process of applying human corrections to a machine-translated draft. MTPE combines the accuracy of human translation with the speed of machine translation. The more human corrections the AI 'brain' learns from, the more it will be able to automatically edit certain types of errors in a machine-translated draft.
Types of Errors:
- Deceptively fluent but incorrect translations
- EXAMPLE: The machine translation might have the sentence, "The boys watch the pigs eat in the field." This sentence is grammatically correct. However, the original text actually said, "The boys eat while watching the pigs in the field." Even though the translated sentence sounds fluent and correct, the verb 'eat' is connected with the wrong actor in the machine translation.
- Inconsistent or non-compliant terminology
- EXAMPLE: The machine translation might have the sentence, "She shaved the sheep." But 'shaved' is the incorrect word for cutting the wool off of a sheep. The correct sentence would be, "She sheared the sheep."
- Repetitions - 'neural babble'
- EXAMPLE: Sometimes in the machine translation, there will be strings of repeated words or phrases where there should not be.
- Omissions - missing words or information
- EXAMPLE: The machine translation might have the sentence, "He picked up the book and began to read." However, the original text says: "He picked up the book covered in markings and began to read." Part of the meaning shared in the sentence was omitted.
- Mistranslations of idioms or figurative language
- EXAMPLE: A common idiom in English is, "It's raining cats and dogs." The machine translation may try to translate that idiom literally in another language, and it may be something like this: "Cats and dogs were falling like water out of the sky."
- Incorrect translations of identifiers (names)
- EXAMPLE: A machine translation will need to be trained to substitute forms of proper names that fit the language of translation. For example, a translation team may prefer that the name 'Yusuf' should be translated as 'Joseph' instead of keeping the original spelling.
- Incorrect capitalization & punctuation
- EXAMPLE: Proper names may not be capitalized, or punctuation may be missing
- Incorrect register and forms of addressing persons
- EXAMPLE: In some languages, a more formal register may be expected to be used in different situations in the text. The machine translation may not pick up on this right away.
What guidelines should I follow when editing a Machine Translation (MT)?
Here are some suggestions to keep in mind while editing a Machine Translation:
