Distance between error distributions vs. benefit from using target priors compared to priors averaged on other language data. Relative advantage reflects the relative improvement that we get between priors estimated on target data vs. other data.
| Articles . | ||
|---|---|---|
| Language . | Weighted L2 dist. . | Target vs. other priors: rel. advantage . |
| Greek | 83.55 | 5.31% |
| Russian | 43.37 | 1.38% |
| Verb agreement | ||
| Language | Weighted L2 dist. | Target vs. other priors: rel. advantage |
| Japanese | 126.98 | 15.38% |
| Thai | 49.09 | 1.04% |
| Articles . | ||
|---|---|---|
| Language . | Weighted L2 dist. . | Target vs. other priors: rel. advantage . |
| Greek | 83.55 | 5.31% |
| Russian | 43.37 | 1.38% |
| Verb agreement | ||
| Language | Weighted L2 dist. | Target vs. other priors: rel. advantage |
| Japanese | 126.98 | 15.38% |
| Thai | 49.09 | 1.04% |