Iterative Recursive Attention Model for Interpretable Sequence Classification

Proceedings of the 2018 EMNLP Workshop BlackboxNLP: Analyzing and Interpreting Neural Networks for NLP, pages 249–257 Brussels, Belgium, November 1, 2018. c⃝2018 Association for Computational Linguistics 249 Iterative Recursive Attention Model for Interpretable Sequence Classification Martin Tutek and Jan ˇSnajder Text Analysis and Knowledge Engineering Lab Faculty of Electrical Engineering and Computing, University of Zagreb Unska 3, 10000 Zagreb, Croatia {martin.tutek,jan.snajder}@fer.hr Abstract Natural language processing has greatly ben- efited from the introduction of the attention mechanism. However, standard attention mod- els are of limited interpretability for tasks that involve a series of inference steps. We de- scribe an iterative recursive attention model, which constructs incremental representations of input data through reusing results of pre- viously computed queries. We train our model on sentiment classification datasets and demonstrate its capacity to identify and com- bine different aspects of the input in an easily interpretable manner, while obtaining perfor- mance close to the state of the art. 1 Introduction The introduction of the attention mechanism (Bah- danau et al., 2014) offered a way to demystify the inference process of neural models. By assigning scalar weights to different elements of the input, we are able to visualize and potentially understand why the model made the decision it made, or discover a deficiency in the model by tracing down a relevant aspect of the input being overlooked by the model. Specifically in natural language processing (NLP), which abounds with variable-length word sequence classification tasks, attention alleviates the issue of learning long-term dependencies in recurrent neu- ral networks (Bengio et al., 1994) by offering the model a glimpse into previously processed tokens. Attention offers a good retrospective explanation of the classification decision by indicating what parts of the input contributed the most to the deci- sion. However, in many cases the final decision is best interpreted as a result of a series of inference steps, each of which can potentially affect its polar- ity. A case in point is sentiment analysis, in which contrastive clauses and negations act as polarity switches of the overall sentence sentiment. In such cases, attention will only point to the part of the input sentence whose polarity matches that of the final decision. However, unfolding the inference process of a model into a series of interpretable steps would make the model more interpretable and allow one to identify its shortcomings. As a step toward that goal, we propose an exten- sion of the iterative attention mechanism (Sordoni et al., 2016), which we call the iterative recursive attention model (IRAM), where the result of an attentive query is nonlinearly transformed and then added to the set of vector representations of the in- put sequence. The nonlinear transformation, along with reusing the representations obtained in previ- ous steps, allows the model to construct a recursive representation and process the input sequence bit by bit. The upshot is that we can inspect how the model weighs the different parts of the sentence and recursively combines them to give the final decision. We test …

Content truncated to 500 words