Abundantia Verborum

3. Tutorial

3.3 Displaying statistics


3.3.5 Filtered diagrams

In section 3.3.3 The display threshold, more precisely in the subsection on Hasse diagrams, we analyzed the correlation between two parameters (groups) on the basis of a single diagram. A more convenient technique is to use the combination of the graph and the filter mechanism, so that each tool takes one parameter into account.

Make sure the workshop "demowork.wrk" is open and active! Now set the members of SEM as displayed labels, choose Schematic as diagram type and set the display threshold to 0% (enabled)! Finally click "OK"! We already know the resulting diagram, which represents the semantic structure of "old". We first saw it when Schematic diagrams were introduced in 3.3.2 Venn, Hasse and Schematic diagrams. This time, let us ask ourselves whether the semantic structure of "old" is systematically influenced by the type of entity it is attributed to.

First set the filter to SAID-OF:(organic entity) and see how this influences the SEM picture! You'll notice that the top and the bottom part of the original schema remains the same. These are the parts we called readings A, B and F in section 3.3.2 Venn, Hasse and Schematic diagrams. The middle part of the original schema, what we used to call readings C, D and E, is gone. Now set the complementary filter NOT(SAID-OF:(organic entity))! This time the original readings C, D and E remain intact, and (apart from a weak relict of reading A) the others have disappeared. We can conclude that the context feature "either or not said of an organic entity" has strong influence on the meaning of "old". Of course, the caveat we've introduced before also applies here. We're comparing a situation of only 21 observations to a situation of only 9 observations. We certainly will need a much larger corpus if we want to corroborate our observation.

The combination of the filter and graph mechanism is the most typical analysis approach in Abundantia Verborum. On the one hand it may be slower than putting everything in one diagram, since it requires trying out two up to a whole series of different filters, but it yields a straightforward, intuitive picture. On the other hand the graph aspect is less flexible than the Boolean operators in filters, but it has the benefit of rendering a more global picture. In summary, filtered diagrams are a good compromise between speed and clarity, because they combine the expressive power of the filter mechanism with the synoptic power of the graph mechanism. To convince yourself, just answer the questions from the above example on the basis of just one of the two mechanisms. You'll notice that you'll either need a lot of queries or a very complex graph.


Back to table of contents