Figure 10

The overview lists the most frequent FP results according to predefined categories. The FP results have again been categorized according to the morphology and the semantics of the missed terms (see Figure 8 above). Again, for all GSCs and the different tagging solutions the counts for the 15 most frequent FP errors are displayed. GP7 and Wh-Ukpmc (GP7) are based on a very large terminological resource that generates BMT and GE FP errors in larger numbers. FP filtering with Chang2 reduces the rates in the case of GP7. The profile of Chang2 differs from the others in the sense that it generates ea-PG and xa-PG FP errors increasing acronyms to larger term structures.