Meaning isn't in words; it isn't in the grammar; language isn't literal.
Take, for example, the innocuous question: "What do you do?"
I've seen this translated literally into Tibetan as: "ག་རེ་བྱེད་ཀྱི་ཡོད།". Now, this obviously misses the mark, and you might be tempted to say it has something to do with the additional "do" that was "not translated".
But this is what's called a "meaningless do". Compare, for example: "How do you do?" to "How are you doing?". The difference in meaning is degree (of formality), but not kind. The meaning is the same: "How are you?".
Whereas "What do you do?" has an entirely different meaning from "What are you doing?" — the first being "What's your occupation?" and the second meaning "What are you up to?".
1.) How do you do? = How are you doing?
2.) What do you do? ≠ What are you doing?
This is a twist on Chomsky's famous example:
1.) John is easy to please = It’s easy to please John
2). John is eager to please ≠ It’s eager to please John
The point is this: Meaning is not encoded in grammatical syntax. Nor is it found in bare vocabulary (see my previous post on how and why). Language is not literal.
Thus, we cannot simply use a "words plus grammar" approach to translate languages (Grammar Translation). "Having all the words" does not ensure accuracy. "Having all the meaning" is what ensures accuracy!
If we cannot assess accuracy by simply (and literally) ensuring that each source word is rendered as a target word, how do we do it?
One method is back-translation. We can translate from target to source to see if meaning is retained. (In an ideal and objective assessment, a second translator unfamiliar with the source text is provided the target, and translates back to the source; the translations are compared for meaning).
In our example, "ག་རེ་བྱེད་ཀྱི་ཡོད།" is back-translated as "What are you doing?" — and when compared with "What do you do?", we can see the mistake...