I think the next thing to do is to start debugging the library. When it decides that the sentence is complete, print it's idea of the complete sentence.
When it parses the data, print each token. It should not take long to determine what the problem is.
All the garbage in there, though, concerns me.