avoid confusion if passing on to udpipe in parallel a data.frame with duplicate doc_id's

[jwijffels]

in that case paragraph_id is within chunks of doc_ids across the cores

[jwijffels]

or if no doc_id's are provided, we get duplicate doc_id's in case of parallel annotation