[pt] Ignore probable English-language toponyms

[p-goulart]

I noticed, in the process of giving LT text from English-heavy Wikipedia articles to evaluate how well we were doing with English word detection, that this was a big weakness.

The issue was that we were generally expecting English phrases (or complete sentences), but these are often single, isolated words. Furthermore, they don't often contain 'frequent' words.

The heuristics here might be improved in the future, but for now I think it strikes a decent balance.

---

We need to note a few things:

• we're not validating if the lower and higher level toponyms match – it'd be cool to check that, but it's not within the scope of the speller, this is just a heuristic to check if some town name is in the English dictionary;
• one thing I dislike about this is that it recognises the town name when it is used right before a state/province/county (e.g. 'Keene, estado de Nova Hampshire' is fine) but not when it's used later (e.g. 'de volta a Keene...');
• I'm wondering if maybe a different tag (_english_toponym_ignore_?) could be used, and then... some magic on the Java side to get all words tagged with _english_toponym_ignore_ and tag them that way even outside the context initially used by the disambiguator... like, dynamically updating our tagger dictionary.