IF I read you correctly, your chief complaint with hierarchies is that they are not responsive to changes over time, but I think that here you’re conflating two aspects of taxonomy systems — the relationship among terms in a lexicon, and the origin of those terms.

Tagging systems aren’t required to be flat and hierarchies aren’t required to be closed. Imagine the construction of hierarchies whose included terms are based on tags created by users.

In this scenario, a content producer (publisher) constructs an initial hierarchy of editorially-chosen tags based on the content of their site and make some decisions about the hierarchy (Is Egypt inside Africa or the Middle East or both?). These publishers can then open the site up to tagging by visitors, and through cloud tags and analytics can track which words are repeatedly being used by site visitors to categorize a particular piece of content. This data can inform the structure of the hierarchy — do most people categorize an item as “Egypt and Africa” or “Egypt and the Middle East”?

Once a tag’s usage has reached a certain threshold, it can be promoted from the folksonomy into the site’s official lexicon and slotted into the hierarchy. Publishers can then use those recently promoted tags to categorize newly created content.

This scheme blends the best of both worlds — it allows content owners to control how their content is being “officially” categorized while still allowing for users to interact and have input.

As for homophones, Google has shown us the power of disambiguation (“Did you mean Britney Spears?”) and through synomyn rings content producers can manage mis-spellings, homophones, and other aspects of language.