In The Preposition Project (TPP), each sense was assigned a semantic relation type by the lexicographer. These types were grouped together into 20 larger classes. The assignment of these two labels was a local decision, that is, without any a priori theoretical perspective. Once completed, the overall collection of these classifications are amenable to more detailed analysis. In particular, each class can be subjected to a digraph analysis and examined in relation to the other classes. In addition, the classes can be compared to the frame element hierarchy. The digraph analysis suggests that several of the classes are really subtypes of other classes. Examination of the frame element hierarchy assists in a clearer perception of the semantic roles filled by prepositional phrases.
Efforts to disambiguate prepositions have been increasing in the last few years, with claims of precision reaching 0.80. All such efforts present results in statistical generalities, with identification of the key factors related to the results. Continued progress in these efforts requires a close examination of limitations that have been noted. In addition, the exploitation of these results requires a close examination of the factors associated with each sense, so that the relevant information for each can be encoded in a meaningful way. This post summarizes the current literature on preposition disambiguation as a prelude to further developments of the data to be encoded in The Preposition Project (see sidebar link).
As described in a previous post, a major step in laying out the frame element (FE) hierarchy involves an examination of the frame element definitions included in the characterizations of each frame in FrameNet. This step has now been completed, at least in a preliminary manner, with the result that the frame element hierarchy has been reduced to 12 primitives and the resulting digraph can now be viewed. The examination of frame element definitions has been both productive and instructive. This post describes how this analysis has proceeded.
As suggested in another post, the assignment of hypernyms to frame elements may introduce additional circularities into the frame element hierarchy. This process is really quite elegant, and illustrates the value of the digraph analysis technique. First, the focus of making assignments is on frame elements that are viewed as primitives, i.e., without hypernyms and without any specific position in the hierarchy. Second, a decision to make a hypernym assignment is strictly local, made without consideration of placement into the hierarchy. Finally, rerunning the digraph analysis takes only a matter of seconds and immediately reveals the fact that an additional circularity has arisen, thus enabling a detailed analysis to break the circularity. This post chronicles the analysis of these additional circularities.