Figure 1.

The NLM-WSD test set and some of its subsets. The 12 terms which Weeber et al. [2] described as "problematic" due to low levels of agreement between annotators are shown in italics. The test set used by Joshi et al. [16] comprises the set union of the terms used by Liu et al. [14] and Leroy and Rindflesch [15] while the "common subset" is formed from their intersection.