Search for: All records

We present an experimental assessment of the impact of feature attribution-style explanations on human performance in predicting the consensus toxicity of social media posts with advice from an unreliable machine learning model. By doing so we add to a small but growing body of literature inspecting the utility of interpretable machine learning in terms of human outcomes. We also evaluate interpretable machine learning for the first time in the important domain of online toxicity, where fully-automated methods have faced criticism as being inadequate as a measure of toxic behavior. We find that, contrary to expectations, explanations have no significant impact on accuracy or agreement with model predictions, through they do change the distribution of subject error somewhat while reducing the cognitive burden of the task for subjects. Our results contribute to the recognition of an intriguing expectation gap in the field of interpretable machine learning between the general excitement the field has engendered and the ambiguous results of recent experimental work, including this study. 

We conduct a large-scale, systematic study to evaluate the existing evaluation methods for natural language generation in the context of generating online product reviews. We compare human-based evaluators with a variety of automated evaluation procedures, including discriminative evaluators that measure how well machine-generated text can be distinguished from human-written text, as well as word overlap metrics that assess how similar the generated text compares to human-written references. We determine to what extent these different evaluators agree on the ranking of a dozen of state-of-the-art generators for online product reviews. We find that human evaluators do not correlate well with discriminative evaluators, leaving a bigger question of whether adversarial accuracy is the correct objective for natural language generation. In general, distinguishing machine-generated text is challenging even for human evaluators, and human decisions correlate better with lexical overlaps. We find lexical diversity an intriguing metric that is indicative of the assessments of different evaluators. A post-experiment survey of participants provides insights into how to evaluate and improve the quality of natural language generation systems. 

We introduce an adversarial method for producing high-recall explanations of neural text classifier decisions. Building on an existing architecture for extractive explanations via hard attention, we add an adversarial layer which scans the residual of the attention for remaining predictive signal. Motivated by the important domain of detecting personal attacks in social media comments, we additionally demonstrate the importance of manually setting a semantically appropriate “default” behavior for the model by explicitly manipulating its bias term. We develop a validation set of human-annotated personal attacks to evaluate the impact of these changes.