benchmarksfactneutralCCG-based parser with directed types achieves 75.9% LF exact match on SLOG, surpassing AM-Parser's 70.8%Machine Learning27 Jul 2026http://arxiv.org/abs/2607.02307v1