Confusion matrix for article errors used in the artificial errors method. Based on the training data from the FCE corpus. The left column shows the correct article. Each row shows the author's article choices for that label and Prob(source|label). The numbers next to the targets show the count of the label (or source) in the data set.
| Label . | Sources . | ||
|---|---|---|---|
| a . | ∅ . | the . | |
| (7,408) . | (44,916) . | (16,719) . | |
| a (7,945) | 0.885 | 0.099 | 0.016 |
| ∅(44,096) | 0.006 | 0.977 | 0.017 |
| the (17,002) | 0.008 | 0.060 | 0.932 |
| Label . | Sources . | ||
|---|---|---|---|
| a . | ∅ . | the . | |
| (7,408) . | (44,916) . | (16,719) . | |
| a (7,945) | 0.885 | 0.099 | 0.016 |
| ∅(44,096) | 0.006 | 0.977 | 0.017 |
| the (17,002) | 0.008 | 0.060 | 0.932 |