4.cuatro Results
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 is one to group (team 0 in both choice) which has the majority of relational adjectives from the standard. Here is the extremely compact group with respect to the clustering traditional.
The brand new discussion is targeted on new class analyses having around three and five clusters just like the kasidie username our very own foundation is actually about three groups (intensional, qualitative, and you may relational) and now we consider all in all, five kinds (earliest groups and additionally polysemous categories: intensional-qualitative and you can qualitative-relational)
Another team (dos from inside the provider A great, 1 in provider B) comes with the majority of qualitative adjectives from the standard, and all the intensional and you can IQ adjectives.
Adjectives that will be polysemous anywhere between a beneficial qualitative and you will an effective relational learning (QR) was strewn by way of all the clusters, even though they tell you a tendency to feel ascribed to the relational cluster in solution B (group 0).
The 5-means answers are portrayed for the Dining table six. For the one hand, the newest desk shows that the 5-way structure discovered from the clustering formula is extremely similar to the three-way structure from inside the Table 5. This is why the three clusters in the A great and you will B provides fundamentally been replicated from the around three earliest groups from inside the C and you will D, correspondingly. Simultaneously, the distinctions amongst the formations acquired playing with theoretical rather than POS have are more obvious throughout the four-ways options. On lay-upwards of experiment, we had expected one team for each classification, plus QR and you may IQ adjectives isolated from inside the a group of their own. This is exactly demonstrably perhaps not borne out in Table 6. What we should select as an alternative would be the fact (a) the brand new combined clusters persist and score saturated in the latest clustering criterion (find groups 0 in service C and 0–1 in solution D, having a mix of Q, QR, and you may R adjectives), and you may (b) a couple of even more short groups are produced (groups step three and you can cuatro in alternatives) with no clear interpretation, suggesting that about three-ways put-up matches ideal the structure exposed by the clustering formula.
Regarding dialogue out of Dining tables 5 and you may 6 we end one the three-way clustering matches the prospective classification much better than the five-way clustering, and that polysemous adjectives aren’t identified as a separate class. Such overall performance suggest that acting polysemous adjectives with regards to more, state-of-the-art kinds isn’t an adequate strategy (we return to this aspect next).
Bear in mind that people defined theoretic and you will POS has actually to compare the fresh formations gotten having fun with theoretically advised and you may theory-independent keeps. Next function analysis, maybe not claimed right here having area grounds, reveals a premier relationship between the extremely detailed attributes of alternatives A and B. step 3 So it shows the newest telecommunications between them feature representations which have admiration into the clustering overall performance: The fresh POS provides elicited because so many discriminative of the clustering formula was precisely people who match the latest theoretic have. So it communications demonstrates to you new resemblance amongst the options obtained into the two types of image as well as the same time frame brings service with the establish concept of the fresh new theoretic has actually.