Reading: Results Tables & Figures — English for Researchers

5. Results (Table 2 caption and discussion)

Table 2: Accuracy (%) on the held-out test set. Bold indicates the best result per row; asterisk (*) indicates statistical significance at p < 0.05 compared to the strongest baseline, via a paired t-test.

Our method outperforms all baselines on 4 of 5 datasets, with the largest margin observed on Dataset-C (+6.2 points). On Dataset-B, the improvement over LRU-K is not statistically significant (p = 0.12), and we report this transparently rather than overstating the result. Figure 3 shows accuracy as a function of training set size; our method's advantage narrows as more training data becomes available, suggesting the gain is most pronounced in low-data regimes. We note that Dataset-E, where our method underperforms the baseline by 1.1 points, differs from the others in having significantly higher label noise — we discuss this limitation further in Section 7.