State-of-the-art spoken language understanding (SLU) models have shown tremendous success in benchmark SLU datasets, yet they still fail in many practical scenario due to the lack of model compositionality when trained on limited training data. In this paper, we study two types of compositionality: (a) novel slot combination, and (b) length generalization. We first conduct in-depth analysis, and find that state-of-the-art SLU models often learn spurious slot correlations during training, which leads to poor performance in both compositional cases. To mitigate these limitations, we create the first compositional splits of benchmark SLU datasets and we propose the first compositional SLU model, including compositional loss and paired training that tackle each compositional case respectively. On both benchmark and compositional splits in ATIS and SNIPS, we show that our compositional SLU model significantly outperforms (up to $5\%$ F1 score) state-of-the-art BERT SLU model.

翻译：最先进的口语理解模型在基准数据集上取得了巨大成功，但由于训练数据有限且缺乏模型组合性，在许多实际场景中仍会失败。本文研究两种组合性：（a）新型槽组合，以及（b）长度泛化。我们首先进行深入分析，发现最先进的口语理解模型在训练过程中常学习到虚假的槽相关性，这导致在两种组合性案例中表现不佳。为解决这些局限，我们创建了首个基准口语理解数据集的组合性划分，并提出了首个组合性口语理解模型，包括分别应对每种组合案例的组合性损失和配对训练。在ATIS和SNIPS的基准数据集及组合性划分上，我们的组合性口语理解模型显著优于最先进的BERT口语理解模型（F1分数提升高达$5\%$）。