Ambiguous Spaces

[Bernd]

run all the analysis twice with options 1 and 2

If you have the resources, I would recommend this and run the analysis both with all and without any spaces. But it is hard to assess without knowing your research goals - not that I am an expert on the matter anyway. The danger of choosing the method giving the most 'sensational' (but not necessarily useful) result is real but as long as you are consistent and make a clear argument for your choice, I see this as a minor problem. Ideally, research should always be focused on rejecting (or at least weighting the parsimony of) hypotheses, not 'proving' own beliefs. Unfortunately this is not the case with the vast majority or VM research as I need not tell you.

One bonus of running 2 analyses is that the differences between input data with and without spaces itself may lead to interesting results.

A good point to research is if uncertain spaces separate common n-grams or they are just ramdom.

This is indeed a good starting point. I admit I have not looked into ambiguous spaces a lot but identifying patterns in their distribution might give some clues about their origin or purpose, if any. But I'd assume this has already been done?

.)are the separated n-grams random?
.)is distribution within lines and on pages random?
.)is distribution across 'languages', 'scribes', 'topics' random?

In any case, the only method to exclude the uncertainty associated with ambiguous spaces is to remove spaces entirely. This of course comes with a cost as Julian said. There's no such thing as a free lunch. But personally this would be my favorite approach.