AMR quality rating with a lightweight CNN
share
Structured semantic sentence representations such as Abstract Meaning Representations (AMRs) are potentially useful in a variety of natural language processing tasks. However, the quality of automatic parses can vary greatly and jeopardizes their usefulness. Therefore, we require systems that can accurately rate AMR quality in the absence of costly gold data. To achieve this, we transfer the AMR graph to the domain of images. This allows us to create a simple convolutional neural network (CNN) that imitates a human rater. In our experiments, we show that the method can rate the quality of AMR graphs more accurately than a strong baseline, with respect to several dimensions of interest. Furthermore, the method proves to be more efficient as it reduces the incurred energy consumption.
READ FULL TEXT