Accents and robustness

[olzama]

It would add some robustness if we would look up things in the lexicon irrespective of the accents/stress. But that's not really a precise approach as the accents are of course meaningful. It's just that people skip them a lot (especially learners). Maybe for a special learner config? But it would cost some performance and the lookup would then be trickier and the chart would be bigger.