The five-way results are depicted in Table 6. On the one hand, the table shows that the five-way structure found by the clustering algorithm is very similar to the three-way structure in Table 5. This means that the three clusters in A and B have basically been replicated by the three first clusters in C and D, respectively. On the other hand, the differences between the structures obtained using theoretical versus POS features are more obvious in the five-way solutions. From the set-up of the experiment, we had expected one cluster per class, plus QR and IQ adjectives isolated in a cluster of their own. This is clearly not borne out in Table 6. What we find instead is that (a) the mixed clusters persist and score high in the clustering criterion (see clusters 0 in solution C and 0–1 in solution D, with a mixture of Q, QR, and R adjectives), and (b) two additional small clusters are created (clusters 3 and 4 in both solutions) with no clear interpretation, suggesting that the three-way set-up matches better the structure uncovered by the clustering algorithm.

Table 6 

First model: Five-way solution contingency tables. Information presented as in Table 5.

.	C: Theoretical .					D: POS .					.
Cluster .	0 .	1 .	2 .	3 .	4 .	0 .	1 .	2 .	3 .	4 .	Total .
I	0	0		0	0	0	0		0	0	2
IQ	0	0		0	0	0	0		0	0	1
Q	7	4		4	2	3	7		2	3	52
QR		3	3	0	0		1	2	1	1	11
R	12		1	0	1		9	5	7	3	35
TotalGS	24	28	42	4	3	20	17	47	10	7	101
Totalcl	857	854	1462	156	192	828	406	1,754	275	258	3,521

.	C: Theoretical .					D: POS .					.
Cluster .	0 .	1 .	2 .	3 .	4 .	0 .	1 .	2 .	3 .	4 .	Total .
I	0	0		0	0	0	0		0	0	2
IQ	0	0		0	0	0	0		0	0	1
Q	7	4		4	2	3	7		2	3	52
QR		3	3	0	0		1	2	1	1	11
R	12		1	0	1		9	5	7	3	35
TotalGS	24	28	42	4	3	20	17	47	10	7	101
Totalcl	857	854	1462	156	192	828	406	1,754	275	258	3,521

View Large