### Table 2: Data for Charles county.

"... In PAGE 9: ...g., Table2 whichistypical#29, we see that this is where much of the computation in both of the r-trees takes place. The advantage of the pmr quadtree over the r-trees in this respect is as high as several orders of magnitude.... ..."

### Table 1: Comparison of previous experiments. R/G=Rubenstein and Goodenough, M/C=Miller and Charles, Res=Resnik, Fin=Finkelstein, Gur=Gurevych, Z/G=Zesch and Gurevych

2006

"... In PAGE 2: ... This experiment was again replicated by Resnik (1995) with 10 subjects. Table1 summa- rizes previous experiments. A comprehensive evaluation of SR measures re- quires a higher number of word pairs.... ..."

Cited by 3

### Table 2: Data for Charles county. 2.0

"... In PAGE 9: ...g., Table2 which is typical), we see that this is where much of the computation in both of the r-trees takes place. The advantage of the pmr quadtree over the r-trees in this respect is as high as several orders of magnitude.... ..."

### Table 6.5: Replica of the Rubinstein and Goodenough and the Miller and Charles experiments

2005

### Table 1: Comparison of semantic similarity measures using the Miller and Charles data

### Table 2: For di erent network sizes, N, we tabulate the number, Nd, of di erent zero-noise limit- cycle attractors, characteristic (or most common) cycle length (char(L( = 0)), characteristic transient (char(Wn?1( = 0)), and average cycle length with non-zero noise ( lt; L gt; ( = 0:01)). Each error bar is the R.M.S. variation from the mean divided by the square root of the number of initial-conditions. At each value of , we observe each net for tmax s

"... In PAGE 17: ... 4.4 Net Size Dependence of Limit-Cycle Period and Number of Di erent Attractors From Table2 , for several di erent network sizes (N 50) with M = 10 without slow threshold noise ( = 0), we have found only a handful of di erent limit-cycles (Nd 6) for each network, without evidence for a systematic dependence on network size. The limit-cycle periods for these networks at zero noise are dominated by one or two di erent periods (char(L)), and also have no systematic dependence on network size.... In PAGE 17: ... We have found however that the transients (char(Wn?1)) prior to convergence to a limit-cycle tends to grow very rapidly with network size N. From Table2 , the mean value Lave of the cycle length grows nearly exponentially with N (Fig. 10), which is similar to N utzel apos;s (1991) exponential scaling of period with system size for coupling symmetries 0:5, and similar to K urten apos;s (1988a) chaotic-phase nets.... ..."

### Table 3: Lake Charles dataset: extraction of terrains at constant error, on the left, and at variable error, on the right.

1997

Cited by 71

### Table 3: Comparison of correlation values for the different measures using the Miller and Charles data

"... In PAGE 5: ...hat this measure performs well for WordNet 1.7.1 . This could be explained by the substantial improvement in the newest version of WordNet, including more uniform distances between words. Table3 shows that it is difficult to replicate accurately experiments that use WordNet-based measures. Budanitsky and Hirst (2001) repeated the Miller and Charles experiment using the WordNet similarity measures of Hirst and St-Onge (1998), Jiang and Conrath (1997), Leacock and Chodorow (1998), Lin (1998) and Resnik (1995).... ..."