Philip Ogren wrote: > Thank you for the feedback. I do need this functionality. It is very > useful for feature extraction. In fact, to me it seems so core to how > feature extraction is usually done that I am surprised that there isn't > a default index(es) for annotation order. Whether you are doing POS > tagging, WSD, NER, chunking, semantic role labeling or whatever you > almost always want to know what the nearby annotations are (e.g. what is > the POS of the token two to the left of the current annotation.) >> This approach is fine, if you need that kind of functionality. I've >> seen it used elsewhere. >> >> --Thilo >> >> >>
Oh, hang on, maybe I didn't understand your note right. You certainly can move an iterator back and forth. UIMA index iterators have more functionality than your regular Java iterator. You can do a moveToPrevious() on the iterator. You can also clone your iterator and move the clone about, if you don't want to move your original. UIMA iterators are more like C++ STL iterators (i.e., pointers) in that way. Does that make sense? --Thilo
