4.4 Show
The contingency tables of the clustering results with three clusters are depicted in Table 5. Part A of the table depicts the solution obtained with theoretical features, while Part B represents the solution obtained with POS features. Rows are gold standard classes and columns are clusters, labeled with the cluster number provided by the algorithm. The ordering of the cluster numbers corresponds to the quality of the cluster, measured in terms of the clustering criterion (see Equation (2)), 0 representing the cluster with the highest quality. In each cell Cij of Table 5, the number of adjectives of class i that are assigned to cluster j by the algorithm is given. The largest value for each class is highlighted (see gray cells).
First model: Three-way solution contingency tables for theoretical and POS features. Rows are gold standard classes, columns are clusters. Row TotalGS shows the number of Gold Standard lemmata and row Totalcl the total number of lemmata contained in each cluster. Note that the column labeled Total represents the row sum for each part (as the number of items per class is identical).
There clearly was one to party (people 0 in both solutions) who has the majority of relational adjectives on gold standard. This is the very lightweight cluster according to the clustering traditional.
This new conversation concentrates on the fresh cluster analyses which have about three and you will four groups because the the basis was three classes (intensional, qualitative, and you will relational) and now we believe a total of five groups (very first categories plus polysemous classes: intensional-qualitative and you can qualitative-relational)
Several other people (2 from inside the solution A beneficial, 1 in solution B) comes with the most of qualitative adjectives in the gold standard, plus the intensional and you can IQ adjectives.
Adjectives which can be polysemous anywhere between a qualitative and a great relational reading (QR) is actually scattered compliment of all groups, even though they show a tendency to getting ascribed to your relational class inside the service B (group 0).
The 5-means results are portrayed for the Desk six. To the one hand, the new table signifies that the 5-means design discover from the clustering formula is extremely like the 3-method build inside the Table 5. Because of this the three groups within the A beneficial and you may B keeps essentially already been replicated because of the about three first clusters from inside the C and you can D, respectively. Concurrently, the distinctions involving the structures gotten having fun with theoretic as opposed to POS keeps be much more apparent about five-method possibilities. Throughout the lay-right up of your test, we’d questioned one to team for each classification, as well as QR and you will IQ adjectives isolated inside a cluster of their very own. This can be certainly not borne in Desk 6. Whatever you select as an alternative is the fact (a) the latest mixed groups persist and you will get packed with the brand new clustering standards (look for groups 0 inside solution C and you will 0–1 in provider D, having a variety of Q, QR, and you can R adjectives), and you will (b) one or two additional quick groups are designed (clusters 3 and you may 4 both in possibilities) and no clear interpretation, suggesting that around three-ways place-right up matches top the dwelling exposed by the clustering formula.
About dialogue off Tables 5 and you can six we finish that the 3-method clustering fits the target group much better than the 5-method clustering, which polysemous adjectives aren’t recognized as a special class. Such performance recommend that modeling polysemous adjectives regarding additional, state-of-the-art kinds is not a sufficient means (i return to this time next).
Keep in mind that people discussed theoretical and you will iamnaughty log in POS has examine this new structures gotten using theoretically informed and concept-separate possess. Next element study, perhaps not claimed right here to have space causes, reveals a top relationship between the extremely detailed top features of alternatives A and B. 3 That it highlights the fresh interaction among them element representations having admiration into clustering abilities: The new POS provides elicited because so many discriminative by the clustering algorithm are truthfully those that correspond to the fresh new theoretic possess. That it communication teaches you the new resemblance involving the choices acquired to the two types of image and at the same time frame provides help towards the expose definition of the brand new theoretical has actually.