Table 6 gives log-likelihood values of the classifier for the training and test sets, along with the upper and lower bounds. The upper bound is the log-likelihood of the model that uses the exact distribution from the Turkers. The lower bound is the log-likelihood of a model that uses only the overall rate of each class in our annotations for the training data.
Log-likelihood values for the training and test data.
| . | Train . | Test . |
|---|---|---|
| lower-bound | − 10813.97 | − 1987.86 |
| classifier | − 8021.85 | − 1324.41 |
| upper-bound | − 3776.30 | − 590.75 |
| . | Train . | Test . |
|---|---|---|
| lower-bound | − 10813.97 | − 1987.86 |
| classifier | − 8021.85 | − 1324.41 |
| upper-bound | − 3776.30 | − 590.75 |