JSON, REST, RDFa, XML, RSS2, ATOM, ATOMPub, OpenSearch, OWL, SRU, csv, Dublin Core plus - what's your poison?
More specifically, what's your institution a) producing and b) consuming? Or, as a developer, what's easiest for you to use?
Tom's made a useful distinction between formats for modelling and for outputs in the comments below. This might resolve some of the 'heavyweight vs lightweight' issues - go as intense as you like on the backend, but start with lightweight outputs on the frontend.
Common RDFa publishing mistakes
"This page contains a number of RDFa publishing mistakes that web authors make when writing RDFa web pages."
Twitter: "We've taken the first steps toward introducing versioning into the Twitter REST API. With a versioned API we can make ambitious improvements *today*, not tomorrow, without worrying about breaking backwards compatibility. This will lead to both a better and more reliable API."
http://groups.google.com/group/twitter-api-announce/browse_thread/thread/2b70bd6ea4aec175?pli=1