Twitter Grammar vs. Annotations

[c1sc0]

I was wondering what your ideas are on embedding data in the tweet
messages themselves vs. in annotations. What kind of info belongs
where? Am I going to use an annotation for something like "I went
#swimming today, the water was #20degrees" or "I had a great pepperoni
at Papa's Pizza"?

Some cases against annotations:

1) Makes you rely on the twitter platform, encoding data in tags stays
portable (facebook, delicious, flickr).
2) For simple single values the whole namespace:key:value may be
overkill?
3) Ease of understaning: even if my client doesn't speak your
particular annotation dialect, my brain can still decode something
like "I drank a huge German #beer:1l"

Some cases FOR annotations:

1) I'm very excited about this, it just may make the whole Semantic
Web thing accessible to normal people!
2) When you look at the avg. size of a microformat, that is well
within the proposed 2K limit. So even in 2K you can put a lot of meta-
info.
3) Annotations may help to clean up the content of twitter msgs.
Twitter grammar & hashtags are approaching the limits of
expressiveness.

Some open questions:

1) What about arbitrary data structures? Will I be able to nest JSON
inside the value?
2) What's it going to be now? Url-parameters? Xml? Json?
3) Has anyone been working on 'translating' microformats 1-to-1 to
twitter annotations? Is that possible (namespace/key constraints)?

I think there will still be a place for 'twitter grammar', but

[c1sc0]

okay, this is silly, but I hit 'send' without finishing that last
sentence, sorry.

I think there will still be a place for 'twitter grammar', but the
more complex data should go in annotations. The question is: how do we
define 'complex'? What do you think?