Correlating neural and symbolic representations of language
This work addresses the need for better interpretability tools in NLP as deep learning becomes dominant, though it is incremental in building on existing analysis techniques.
The authors tackled the problem of understanding neural language models by developing two methods to quantify the alignment between neural activation patterns and symbolic syntax trees, demonstrating expected results on a synthetic language and applying them to English sentences.
Analysis methods which enable us to better understand the representations and functioning of neural models of language are increasingly needed as deep learning becomes the dominant approach in NLP. Here we present two methods based on Representational Similarity Analysis (RSA) and Tree Kernels (TK) which allow us to directly quantify how strongly the information encoded in neural activation patterns corresponds to information represented by symbolic structures such as syntax trees. We first validate our methods on the case of a simple synthetic language for arithmetic expressions with clearly defined syntax and semantics, and show that they exhibit the expected pattern of results. We then apply our methods to correlate neural representations of English sentences with their constituency parse trees.