Speaker: Ishita Dasgupta (MIT)
Title: Evaluating Compositionality in Sentence Embeddings
Date and time: Monday, March 3, 5:00-6:00pm
Location: 46-5156
Abstract:
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 some of the decision rules learned by InferSent and find that they are largely driven by 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 NLP systems.