### Table 1: Translation accuracy of selected n-gram model automatically trained on different sources of training data Training Source Top 1 Top 5 Top 20

2003

"... In PAGE 3: ... We started with the assumption that it was best to use a variety of sources for training, so that more spelling variations might be covered. Table1... ..."

### Table 1. Effect of language model training data

2004

"... In PAGE 2: ... As a result, we get 30820 word-pronunciation entries (24437 distinct words), for which a trigram model is trained. The effect of training data size is clearly confirmed in Table1 . WER (Word Error Rate) is significantly reduced according to the increase of the data.... ..."

Cited by 3

### Table 1: Effect of language model training data

"... In PAGE 2: ... Here, we adopt a simple trigram model of word-pronunciation entries. The effect of training data size is clearly confirmed in Table1 . WER (Word Error Rate) is significantly re- duced according to the increase of the data.... ..."

### Table 3. Recall (R) and Precision (P) Values for the Models Generated for Each Target (Training) Considering the VA Seta

2005

"... In PAGE 4: ... Both methods show recall and precision values either close or equal to 100%. As regards training, both methods perform equally well on the data set with a small number of actives ( Table3 ) and on the data set with a larger number of actives (Table 4). Tables 3 and 4 include, as well, average and standard deviations for 25 randomly selected data sets for each target in the case of SVM.... ..."

### Table 12.49: Effect of language model size on translation quality (NIST mteval score) for large data track.

### Table 4. Effect of alignment model on translation quality.

2000

"... In PAGE 12: ... More refined alignment models do not only result in improved alignment quality, but also improved translation quality. The effect on the translation quality is shown in Table4... ..."

Cited by 7

### Table 3. Effect of Alignment Model on Translation Quality.

2000

Cited by 7

### Table 6. Effectiveness Analysis for Selection of Interventions

"... In PAGE 16: ...16 which omits access to electricity.26 This conjecture is borne out by the comparison of the two specifications (mod-6 and mod-2) as shown in Table6 . We find that, when not controlling for all other variables, the estimated gross income effect is large (-70.... In PAGE 17: ... In the effectiveness analysis, we present results using both linear and nonlinear specifications 28. As shown in Table6 , the estimated effectiveness coefficients from the nonlinear model are broadly in line with that from the linear model, except for income (with GDP growth rate being doubled in the nonlinear model) and access to sanitation. Table 6 summarizes the estimated cost-effectiveness coefficients for various interventions to avert one infant and one child death per 1000 births.... In PAGE 17: ... As shown in Table 6, the estimated effectiveness coefficients from the nonlinear model are broadly in line with that from the linear model, except for income (with GDP growth rate being doubled in the nonlinear model) and access to sanitation. Table6 summarizes the estimated cost-effectiveness coefficients for various interventions to avert one infant and one child death per 1000 births. In the following discussion we focus only on statistically significant estimates from the nonlinear model.... In PAGE 17: ... According to WDI, the annual per capita GDP among LDCs between 1990-1999 is less 1%. However, it should be emphasized that figures presented in Table6 should be viewed as an illustration of the methodology, and not be taken too literally for policy purposes. The main reason is rooted in the limitations in cross-country data, as revealed in the econometrics results presented in the above section, in providing robust empirical 28 We chose semi-log function so that the rate of mortality reduction of any intervention depends on the level of mortality and the cost-effectiveness coefficients.... ..."

### Table 8. Fraud Data Set, Categorical Data, Logistic Regression

"... In PAGE 6: ... This yielded a very good correct classification rate, as only 22 of 1000 test cases were fraudulent. Table8 gives results for the logistic regression model. Table 8.... ..."

Cited by 1