This is somewhat similar to the language vector embedding, isn't it? And the article asks the reasonable question "what is the difference between having a sing

This is somewhat similar to the language vector embedding, isn't it? And the article asks the reasonable question "what is the difference between having a single word for a thing versus a commonly understood cluster of words?". It's not a hard boundary. Every translation loses a little bit of information but potentially brings in different connotations. The things that translators and localizers argue about endlessly: do we look for the words that most closely match the other words, or do we look for feeling and meaning that most closely matches the original intent?