Yes, you are absolutely correct! When dealing with HTML entities this process should be done in two steps.

Step 1: Data extraction - Use an HTML Parser to pull out all of the data first, so that it can be represented in a humanly readable format.

Step 2: Sentence extraction - Use your sentence parser to break the humanly readable information up into separate sentences.

The problem becomes even more exacerbated when you have to also consider different tagging formats such as XML and its many variants, or an SGML standard, etc. etc. ad. nauseum.

Here is another thought I had recently. Would it be possible to write a Grammar and use RecDescent to pull out sentences? I really haven't investigated it thoroughly yet, but I thought it might be an interesting exercise.