### Table 1. Simple programs.

2005

"... In PAGE 6: ...Simple Programs We start demonstrating the feasibility of our approach with some examples taken from papers [12, 17]. Table1 shows the example programs with the corresponding number of rules applied in the KeY system. Note that l, h, and r are single... In PAGE 10: ...2 Examples In this subsection we show how this approach performs on the example programs from Section 3. On the simple programs in Table1 the number of applied rules increased, but in most cases not more than linearly. No user interaction was needed.... ..."

Cited by 47

### Table 3.1: Computation time for elimination and back substitution for matrices of size n by n. From table 3.1 it is seen that solving a linear system of equations with 100 unknowns is a simple task, however a full 1000 1000 matrix is already near the limit of what can be solved at a reasonable cost on a small computer (though this is changing quite rapidly). The main point is that the time to solve the equation is going up much quicker than the size ofthe problem (size O(n2).

### Table 1: Some simple linear models; Full Sample ( n = 34; t-statistics in parenthesis)

"... In PAGE 4: ... of .17. This relationship, moreover, does not disappear if we control for the variables usually considered in empirical studies of the relationship between election laws and party fragmentation -- the magnitude of legislative electoral districts ( D ), the presence or absence of adjustment districts ( AD ), and societal heterogeneity ( H ) -- or if we use the more common measure of party fragmentation, the `effective apos; number of parties, EN . 3 As Table1 shows, a directly elected 2 The data analysis presented in this paper was conducted using the SST package. In this and all subsequent regressions we computed heteroskedasticity-consistent standard errors using White =s (1980) method.... In PAGE 5: ...president has its own positive and highly statistically significant influence on the number of parties that compete for legislative seats. Of course, the number of observations in Table1 is not great ( n = 34). And since most countries in our sample are represented by more than one observation, one could suppose that the statistical pattern portrayed there can be influenced by the idiosyncratic characteristics of countries (i.... In PAGE 5: ... ), and endogenous institutional selection (i.e., an existing proto-party system that influences the choice of presidential versus parliamentary governmental forms). In fact, though, the coefficients for P in Table1 are relatively robust. For example, - If, to eliminate the country-specific effects allowed by multiple observations from each country, we take only one election per country (the second, because gt;founding = elections are considered suspect due to their plebiscite nature) we get N = 6.... In PAGE 14: ...lsewhere, in fact, we show that for established democracies (e.g., Canada, the United States, Germany, Finland, England, Australia, Netherlands, Iceland, and so on), this model yields a better fit than does one that merely incorporates H as a linear additive variable or that ignores H altogether (Ordeshook and Shvetsova 1994). But here we see that although the coefficients for H ln( D ) are statistically significant (except when our dependent variable concerns the number of parties winning seats), such a model for our complete sample ( Table1 ) is no better in terms of adjusted R 2 than one that ignores heterogeneity altogether, and is strictly worse when we restrict the analysis to Central Europe (Table 2). There is no indisputable explanation for this pattern, but the most evident hypothesis is that when democratic process is suddenly thrust upon a society, as it was in so many of the countries in our sample, parties designed to `serve apos; or otherwise take advantage of basic social cleavages coexist with the other parties that form at this time, and that total number greatly surpasses what can be sustained in equilibrium as a product of the combined influence of a polity apos;s electoral laws and social cleavages.... In PAGE 15: ... For another comparison with equivalent implications, consider Lijphart apos;s (1990) regime data, which also pertains to established democracies. Regressing `effective apos; number of parties against the log of average district magnitude yields the equation (see Table1 in Ordeshook and Shvetsova 1994) EN = 3.24 + .... In PAGE 16: ... Thus, as we subsequently argue, we cannot reject the hypothesis that the coefficient for D is zero in presidential systems. 12 These numbers come from the first regression in Table1 after setting P = 0 and D = 25, since 25 is the maximum district magnitude in our sample. 13 Notice the slight positive slope of the line for presidential systems since in this regression ... ..."

### Table 2: Performance on accept/reject classification and the top-level task, on 12 different configurations of the system. Threshold = simple threshold on average confidence; Linear = SVM classifier with linear kernel; Quad = SVM classifier with quadratic kernel; Quad/r = recalibrated version of SVM classifier with quadratic kernel; A = in-domain and correct semantic interpretation; B = in-domain and incorrect or no semantic interpretation; C = out-of-domain; All = standard classifier error rate over all data; u2 = weighted average of classifier error using u2 weights; Task = normalised task metric score; j = value of svm-light J parameter.

### Table 8.2: Word error rates (%) on clean RM Feb apos;89 test set for various parameterisations In order to use PMC to compensate the models it was necessary to alter the speech parameter set from that used in the standard HTK RM recogniser. To test the e ect of this, a series of systems were built using a variety of speech parameters and tested on the February 1989 test set. The results are shown in table 8.2. The rst column shows the type of energy term used, Log is normalised log-energy, Cepstra is unnormalised C0. The second column is the form of dynamic coe cients, Regr. is linear regression based, and Di . is simple di erences. These apply to both the delta and delta-delta parameters. There was a slight di erence in performance when changing the parameter set, par- ticularly for the single mixture component case. However these di erences were greatly reduced in the two mixture component case. From these results, it appears that there is little di erence in clean performance between using log-energy with linear regression as used in the standard system and the zeroth Cepstra with simple di erences used for the PMC experiments.

### Table 2. Simple service time approximation

2002

"... In PAGE 12: ... Table 1 shows the maximum through- put and the corresponding service time for each URL size. Table2 compares the measured service times to ser- vice times computed using the linear approximation Ts(x) = A + Bx where x is URL size, A = 1:604, B = 0:063. The constant A can be thought of as the time it takes to serve a zero-size URL.... In PAGE 12: ... In order to model service time more accurately we use a composition of two linear approximations, one es- timates service time if the end-system is the bottleneck and the other estimates service time if network band- width is the bottleneck. While the former is given as be- fore by Ts(x) = 1:604 + 0:063x, Table2 suggests that the latter be given by Ts(x) = 0:093x, which is equiv- alent to stating that the network saturates at a transfer rate of approximately 86Mb=s. We then take the larger of the two service times to account for the bottleneck resource.... ..."

Cited by 131

### Table 2. Simple service time approximation

2002

"... In PAGE 12: ... Table 1 shows the maximum through- put and the corresponding service time for each URL size. Table2 compares the measured service times to ser- vice times computed using the linear approximation T s (x) = A + Bx where x is URL size, A = 1:604, B = 0:063. The constant A can be thought of as the time it takes to serve a zero-size URL.... In PAGE 12: ... In order to model service time more accurately we use a composition of two linear approximations, one es- timates service time if the end-system is the bottleneck and the other estimates service time if network band- width is the bottleneck. While the former is given as be- fore by T s (x)=1:604 + 0:063x, Table2 suggests that the latter be given by T s (x)=0:093x, which is equiv- alent to stating that the network saturates at a transfer rate of approximately 86Mb=s. We then take the larger of the two service times to account for the bottleneck resource.... ..."

Cited by 131

### Table 5 Simple Transition System

"... In PAGE 11: ... 4.1 Untyped Transition Semantics The untyped transition semantics is given in Table5 . It is worth noting that almost all the reduction rules explicitly mention a context containing an am- bient, except for the rule for local communication.... ..."