Another project last year did that by using the "Visual Placements" question. However, the number of words to be marked is quite limited (at least if you want them in the described data format).
Another option would be the (rather new) question typ "SVG selection". If you created the text as SVG image (e.g. using InkScape) and make each word a separate object, then the respondents would be able to select single words. All of them, if they liked. Although this would work quite fine, I assume it will be some work to create the SVG.