It's often more like this:
1) Spider downloads pages and parses for links. The pages are sent to...
2) Another app stores them.
3) A third app parses the stored pages and extracts key words and phrases for storage in a database that will be accessed by...
4) A fourth app that queries the database based on keyword searches.
It's of course even more complex than that because there are so many other factors these days. Steps three and four branch out into a lot of sub processes and step four often retreives bits of the data stored in step 2.
Some day (maybe not so far off) computers and storage will be fast enough to do away with steps 2 and 3 altogether.
Semantics, as I understand the term, happens in steps three and four where additional information is stored/interpreted about the context that each keyphrase is found in. This could be the surrounding paragraph, the theme of the page, the presence of certain phrases with a high statistical probabability of classifying nearby text in a certain topic, etc...
Any search engine that's trying to do this sort of thing is changing their algoritms so frequently that I don't think it's really something you can white hat optimize for at this point.
In fact, like everything else in search engine algorithms, if it ever does become possible to optimize for it then it will quickly most of it's value.
Right now I'd say it's safe to assume that if your content is legitmate and makes sense then semantic ranking will benefit you, whereas if your content is disjointed (i.e. scraped) or mixes too many topics in a small amount of text then it might hurt you.