Language acquisition and translation involves mapping between systems of meaning, whether between percepts of different sensory modes, between direct experience and symbols, or between sets of verbal encodings.
One aspect of such mappings is that the boundaries and clustering of meanings in each system may be quite different. In human languages this is most apparent in prepositions, in systems of homonyms and in alternative meanings for individual words.
The existence of homonyms, ambiguity and multiple connotations and meanings for individual words reflects the predicament of encoded language in general: a finite lexicon of discrete words is required to represent a continuous field of infinite meanings. In order to speak in words and sentences of manageable length, with a finite vocabulary, individual words must carry multiple meanings.
Ambiguity is intrinsic to language.
Over the course of time, as more sentences are translated and re-entered into the database, the translation algorithm becomes "smarter" and better able to resolve ambiguous words.