Evaluating Compositionality in Sentence Embeddings
This work addresses the problem of evaluating and improving compositional understanding in AI systems for researchers, though it is incremental as it builds on existing methods with a new dataset.
The authors tackled the challenge of compositional semantics in AI by creating a new natural language inference dataset requiring compositionality, and found that state-of-the-art sentence embeddings performed poorly on it, with performance improving through dataset augmentation without harming original training performance.
An important challenge for human-like AI is compositional semantics. Recent research has attempted to address this by using deep neural networks to learn vector space embeddings of sentences, which then serve as input to other tasks. We present a new dataset for one such task, `natural language inference' (NLI), that cannot be solved using only word-level knowledge and requires some compositionality. We find that the performance of state of the art sentence embeddings (InferSent; Conneau et al., 2017) on our new dataset is poor. We analyze the decision rules learned by InferSent and find that they are consistent with simple heuristics that are ecologically valid in its training dataset. Further, we find that augmenting training with our dataset improves test performance on our dataset without loss of performance on the original training dataset. This highlights the importance of structured datasets in better understanding and improving AI systems.