Naturalness and user preference percentage of the best performing systems for each pairwise system comparison (winners in bold). Significance was computed using a two-tailed binomial test (*p < 0.05, **p < 0.01, ***p < 0.001).
| System A . | System B . | Nat . | Pref . |
|---|---|---|---|
| FLM n-best | Handcrafted | 52.2 | 52.3 |
| FLM n-best | FLM 1-best | 57.3*** | 54.2* |
| FLM n-best | Perceptron | 51.2 | 52.7 |
| Perceptron | Handcrafted | 54.2* | 54.2* |
| System A . | System B . | Nat . | Pref . |
|---|---|---|---|
| FLM n-best | Handcrafted | 52.2 | 52.3 |
| FLM n-best | FLM 1-best | 57.3*** | 54.2* |
| FLM n-best | Perceptron | 51.2 | 52.7 |
| Perceptron | Handcrafted | 54.2* | 54.2* |