### Table 5 Results of Log-Linear Analysis

1995

"... In PAGE 9: ...the significance of interactions between panel assignment, language of census form and item nonresponse for each of the demographic and housing questions on the census form. Table5 reports our results for the demographic census items and Table 6 for the housing census items. We fit four models treating the demographic and housing variables as response variables.... In PAGE 9: ... The other variables in the models are panel assignment (control, dual forms or bilingual panel) denoted by quot;T quot; and language (Spanish or English) of the census form returned by mail to the Census Bureau, denoted by quot;L quot;. The results of log-linear analysis reported in Table5 and Table 6 were derived using the software package CPLX (Fay 1989). This software program adjusts for the complex sample design of the SFAT by using a jackknife estimation method.... In PAGE 9: ... Demographic Census Items: Results Presented in Table 5 Model 1 includes the interaction between the response variable and panel assignment {RT}, panel assignment and language of census form {TL} and the response variable and language of census form {RL} for population questions. As shown in Table5 , model 2 is similar to model 1, however, it excludes the interaction response by panel assignment term {RT}. This model does not provide a fit of the data for the marital status and date of birth variables.... ..."

Cited by 1

### Table 6 Results of Log-Linear Analysis

1995

"... In PAGE 9: ...the significance of interactions between panel assignment, language of census form and item nonresponse for each of the demographic and housing questions on the census form. Table 5 reports our results for the demographic census items and Table6 for the housing census items. We fit four models treating the demographic and housing variables as response variables.... In PAGE 9: ... The other variables in the models are panel assignment (control, dual forms or bilingual panel) denoted by quot;T quot; and language (Spanish or English) of the census form returned by mail to the Census Bureau, denoted by quot;L quot;. The results of log-linear analysis reported in Table 5 and Table6 were derived using the software package CPLX (Fay 1989). This software program adjusts for the complex sample design of the SFAT by using a jackknife estimation method.... In PAGE 10: ... Using the same rationale, four similar models were estimated for the housing data. The results are shown in Table6 . The p-values for model 1 indicate that it fits the data for most of the housing variables.... ..."

Cited by 1

### Table 2: Summary of Log-linear Models

"... In PAGE 22: ... 1994, Boehm 1981]. Table2 shows the exponent, b, de- rived from regression of different data sets for non-linear models [Banker and Kemerer 1989, Boehm 1981, pp.86].... In PAGE 107: ... Self-descriptive code; documen- tation up-to- date, well-orga- nized, with design rationale. SU Increment to ESLOC 50 40 30 20 10 Table2 0: Rating Scale for Software Understanding Increment SU : Rating Scale for Assessment and Assimilation Increment (AA) AA Increment Level of AA Effort 0 None 2 Basic module search and documentation 4 Some module Test and Evaluation (T amp;E), documentation 6 Considerable module T amp;E, documentation 8... ..."

### Table 6 Extremity probabilities of expert judgements for the Log-Linear function

1999

"... In PAGE 5: ...osterior distribution as well as the mean deviance. Fig. 5 shows the density plots of the parameters. The extremity probabilities associated with each expert judgement are given in Table6 . The judgements of the experts are slightly better accommodated in the Log-Linear model, as the lower extremity probabilities are higher than for the Power Law model.... ..."

### Table 9. Fit of the log-linear model with the interactions: type of

1994

### Table 3 Word error rates (%) obtained using linear and log-linear interpolation of MELMs and the class trigram Method Linear-interpolation Log-linear interpolation

2003

"... In PAGE 17: ...ELM weights were 0.41 for MELM1, 0.64 for MELM2, and 0.40 for MELM3. The results are shown in the left-hand column of Table3 . We did not observe any improvement in WERC213s for linearly interpolated models over the individual MELM-based methods.... In PAGE 18: ...logP1 + logP2, where Pi is the probability assigned by the ith model. The results of log-linear interpolation are shown in the right-hand column of Table3 . Here, we observe that we can achieve a significant reduction in the word error rate using the MELM3 and class trigram mixture over MELM3 alone.... ..."

### Table 2. Comparison of AER for results of using IBM Model 3 (GIZA++) and log-linear models.

"... In PAGE 6: ... In other words, our log- linear models share GIZA++ with the same parame- ters apart from POS transition probability table and bilingual dictionary. Table2 compares the results of our log-linear models with IBM Model 3. From row 3 to row 7 are results obtained by IBM Model 3.... In PAGE 6: ... Our log-linear models still make use of the parameters generated by GIZA++. Comparing Table 3 with Table2 , we notice that our log-linear models yield slightly better align- ments by employing parameters generated by the training scheme 15H5354555 rather than 15H535, which can be attributed to improvement of param- eters after further Model 4 and Model 5 training. For log-linear models, POS information and an additional dictionary are used, which is not the case for GIZA++/IBM models.... ..."

### Table 2. Comparison of AER for results of using IBM Model 3 (GIZA++) and log-linear models.

"... In PAGE 6: ... In other words, our log- linear models share GIZA++ with the same parame- ters apart from POS transition probability table and bilingual dictionary. Table2 compares the results of our log-linear models with IBM Model 3. From row 3 to row 7 are results obtained by IBM Model 3.... In PAGE 6: ... Our log-linear models still make use of the parameters generated by GIZA++. Comparing Table 3 with Table2 , we notice that our log-linear models yield slightly better align- ments by employing parameters generated by the training scheme 15H5354555 rather than 15H535, which can be attributed to improvement of param- eters after further Model 4 and Model 5 training. For log-linear models, POS information and an additional dictionary are used, which is not the case for GIZA++/IBM models.... ..."

### Table 6. The log-linear model (5) (6) Predictor Coef SE Coef T P

2003

"... In PAGE 12: ...943 and with a standard error of 0.11, see Table6 ).... In PAGE 13: ...omoscedastic either. This is observed by comparing the plots of Figure 2 and Figure 4. Compared with the log-linear model (3), model (9) seems no better. In addition, comparing Table6 and Table 7, we observe that the intercept of the log-linear model (3) has a higher t- value than the intercept of model (9), and that the slope coefficients have similar t-values. Overall, the log-linear model seems the best choice.... ..."

Cited by 6

### Table 3: Performance of conditional log-linear models for the parse selection task (accuracy).

2003

"... In PAGE 3: ...ame features. We refer to the log-linear models as LTrigram, LPCFG-1P, and LPCFG-3P. These models correspond to the generative models Trigram, PCFG-1P and PCFG-3P respectively. Table3 shows the accuracy of parse selection using the conditional log-linear models. 3 Discussion It is useful to look at accuracy results from models on the two corpora versions depending on the ambiguity level of sentences.... ..."