From what I have been able to determine from the parser code (hey, about some comments guys?), it does not analyze grammatical structure at all. It looks for non-overlapping occurences of known category words, mostly names of things and applicable verbs. It is not clear if it worries about word order at all.
While this does limit the ability to process fine distinctions, it does make the parser processing pretty language independant.
However, to get high quality speech-to-text, the speech recognizer does need to know about all those other words (particles, prepositions, articles, pronouns) in order to detect word boundaries i the first place. And knowing what word orders are allowed provides useful information about what word orders are not allowed and should be eliminated from possible results. This part is highly language specific.
Early speech recognition research at CMU decades ago showed that recognition accuracy goes up the more the recognizer knows what is, and is not, a valid “utterance”. You can’t just give it a big dictionary of words.