Our title today comes from Korzybski; follow the link to see how he used it, but I'd like to use the metaphor to discuss language, and the study of language. The map—here, language—is not the territory—the experiences, or representative mind-states, we are using that language to communicate.
For a very concrete example, let's consider the English exclamation "ow!" (or "ouch!"), which is used within our linguistic map to demarcate the territory of "a sudden, surprising pain"—which is the actual terrain of experience itself.
In Tibetan, however, our territory of "sudden, surprising pain" is actually split into two districts: ཨ་ར་ (a ra, a sudden dull pain; for example, from being hit or struck) and its adjacent neighbor ཨ་ཚ་ (a tsha, a sudden sharp pain; for example, from being pinched or burning oneself).
If we consider a situation of hypothetical translation, our simple, concrete example becomes suddenly complex. Our English-into-Tibetan must now take note of the context in which the exclamation was uttered (was it a dull or sharp pain?), while our Tibetan-into-English must now concern itself with whether or not giving more context or leaving it absent is the more appropriate choice.
To put it in linguistic terms, vocabulary is not isomorphic (one word in one language does not simply equal one word in another); and Messages contain information that is both linguistic and extralinguistic (contextual information).
This is why leaving information out, or adding extra information in, can be the right decision in translation. Fidelity—how accurate we are—is not measured by whether we use the same grammar or vocabulary (the same linguistic information); it is measured by whether or not we convey the same Message.