The results can also be interpreted this way: for sentences with grammatical syntax, the effect of Semantic Interpretability was marginally significant, estimated as 2.7% lower accuracy for uninterpretable sentences, p ≈ 0.055 ( Figure 4 top, GI versus GU).
For sentences with ungrammatical syntax, there was a nonsignificant trend in the opposite direction, 0.05% percent higher accuracy for uninterpretable sentences (p ≈ 0.76).
Crucially, however, that small effect was itself significantly different (p ≈ 0.032) in comparison to what we observed in sentences with ungrammatical syntax, where there was a numeric trend in the opposite direction (0.5% higher accuracy) that was nowhere near statistical significance (p ≈ 0.76).