Many people naively believe there is a one-to-one correspondence between word meanings in different languages, and that translation is a simple matter of replacing each word in the source language with the corresponding equivalent word in the target language.
In fact language (and translation) is much more complex than that. One reason there is no one-to-one correspondence between word meanings is that words often have different conceptual–lexical mappings in different languages. For example, the Spanish noun reloj covers the concepts denoted by both clock and watch in English. Similarly, the English verb know, which covers both knowing facts and knowing people, maps to two distinct verbs in Spanish: saber for the former and conocer for the latter.
These conceptual–lexical mismatches tend to be even more dramatic when translating back and forth between English and Japanese, which are completely unrelated languages.
A good example is the English verb identify. This verb, which came into English in the mid 17th century from the medieval Latin verb identificare, originally meant ‘treat as being identical with.’ Since then, the word has acquired a large number of additional senses and nuances. Today, a witness can identify a suspect, a CEO can identify problems to be solved, and a teenager can identify with Harry Potter. There is a long list of Japanese verbs that can be translated as identify depending on the context, including 明確化する (clarify), 発掘する (excavate), 抽出する (extract), 位置付けられる (position), 決定する (determine), 把握する (understand), 発見する (discover), 確認する(confirm), 識別する (discriminate), and 割り出す (figure out).
In the other direction, a classic example is the Japanese verb 対応する (taiō suru), which maps to a diverse set of English verbs, including correspond, respond, handle, coordinate, match, support, deal with, and be compatible with. Machine translation systems still have lots of trouble capturing the many senses and nuances of 対応する, which is of course good news for human translators of Japanese.