Perplexity scores

[emarcus255]

Hi.

I am trying to calculate perplexity scores on my data set using marian-scorer and to be honest I am not sure if I am doing it correctly and if the scores returned by marian-scorer are the right one.

marian-scorer -m model/model.npz -v model/spm-src.spm model/spm-trg.spm -t data/source.test.src data/source.test.trg > marian_psppl.score

I know in the --summary parameter I can pass what type of score to get but what actually scores I am getting for each sentence of my data set (without passing the summary option)? The scores I am getting have all negative value what makes me wonder if they are correct.