Procedural Reasoning Networks for Understanding Multimodal Procedures

Abstract

This paper addresses the problem of comprehending procedural commonsense knowledge. This is a challenging task as it requires identifying key entities, keeping track of their state changes, and understanding temporal and causal relations. Contrary to most of the previous work, in this study, we do not rely on strong inductive bias and explore the question of how multimodality can be exploited to provide a complementary semantic signal. Towards this end, we introduce a new entity-aware neural comprehension model augmented with external relational memory units. Our model learns to dynamically update entity states in relation to each other while reading the text instructions. Our experimental analysis on the visual reasoning tasks in the recently proposed RecipeQA dataset reveals that our approach improves the accuracy of the previously reported models by a large margin. Moreover, we find that our model learns effective dynamic representations of entities even though we do not use any supervision at the level of entity states. 1

A Bird’s Eye View of Natural Language Processing and Requirements Engineering

“…Visual question answering is a method that addresses the challenging unimodal aspect of NLP systems [78]. Many other methods are used to integrate multimodality into NLP structures, including declarative learning-based programming [79], multimodal datasets [80], procedural reasoning networks [81], and unified attention networks [82].…”

Latent Alignment of Procedural Concepts in Multimodal Recipes

“…Procedural reasoning has been investigated on different tasks (Amac et al, 2019;Park et al, 2017). While PRN (Amac et al, 2019) is proposed on RecipeQA, their model does not apply to the textual cloze task.…”