Self-Assembling Modular Networks for Interpretable Multi-Hop Reasoning

By Yichen Jiang, Mohit Bansal. 2019

74 citations recorded by the source corpus.

Multi-hop QA requires a model to connect multiple pieces of evidence scattered in a long context to answer the question. The recently proposed HotpotQA (Yang et al., 2018) dataset is comprised of questions embodying four different multi-hop reasoning paradigms (two bridge entity setups, checking multiple properties, and comparing two entities), making it challenging for a single neural network to handle all four. In this work, we present an interpretable, controller-based Self-Assembling Neural Modular Network (Hu et al., 2017, 2018) for multi-hop reasoning, where we design four novel modules (Find, Relocate, Compare, NoOp) to perform unique types of language reasoning. Based on a question, our layout controller RNN dynamically infers a series of reasoning modules to construct the entire network. Empirically, we show that our dynamic, multi-hop modular network achieves significant improvements over the static, single-hop baseline (on both regular and adversarial evaluation). We further demonstrate the interpretability of our model via three analyses. First, the controller can softly decompose the multi-hop question into multiple single-hop sub-questions to promote compositional reasoning behavior of the main network. Second, the controller can predict layouts that conform to the layouts designed by human experts. Finally, the intermediate module can infer the entity that connects two distantly-located supporting facts by addressing the sub-question from the controller.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition, Natural Language Processing, Model Architectures

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Computer Vision and Pattern Recognition

In Natural Language Processing

In Model Architectures

Citations in the corpus

Explainable Multi-hop Verbal Reasoning Through Internal Monologue (2021, background): “Another direction are the neural modular networks, where what components to use are determined dynamically for each question (Gupta et al., 2019; Jiang and Bansal, 2019).”
Unsupervised Question Decomposition for Question Answering (2020, methodology): “Such methods usually require strong, program-level supervision to generate programs, as in visual QA (Johnson et al., 2017b) and on H OTPOT QA (Jiang & Bansal, 2019b).”
Neural Module Networks for Reasoning over Text (2019, neutral): “Concurrently, Jiang & Bansal (2019) apply NMN to HotpotQA (Yang et al., 2018) but their model comprises of only 3 modules and is not capable of performing symbolic reasoning.”
Text Modular Networks: Learning to Decompose Tasks in the Language of Existing Models (2021, background): “1266 compare against formulations of NMNs for HotpotQA (Jiang and Bansal, 2019) and DROP (Gupta et al.”
Reason first, then respond: Modular Generation for Knowledge-infused Dialogue (2021, background): “The approach of text modular networks promises more interpretable answers to multi-hop questions (Khot et al., 2020; Jiang and Bansal, 2019; Gupta et al., 2020).”
Decomposed Prompting: A Modular Approach for Solving Complex Tasks (2022, background): “…for Multi-Step Reasoning Our work follows a long literature in NLP on neural modular modeling architectures (Andreas et al., 2016; Talmor & Berant, 2018; Jiang & Bansal, 2019; Gupta et al., 2020; Perez et al., 2020; Khot et al., 2021; Levine et al., 2022) for question-answering and other tasks.”
Teaching Machine Comprehension with Compositional Explanations (2020, methodology): “This module is similar to the ﬁnd module in (Jiang & Bansal, 2019) in its motivation, but we speciﬁcally design ours to be a ranking based model with discrete boundaries instead of continuous attention weights, so that the output ﬁts in the search algorithm and answering procedure in Sec.”
Obtaining Faithful Interpretations from Compositional Neural Networks (2020, methodology): “Work on NMNs thus far (Yang et al., 2018b; Jiang and Bansal, 2019) has overlooked systematically evaluating faithfulness, performing only qualitative analysis of intermediate outputs.”
Case-Based Abductive Natural Language Inference (2020, background): “The work presented in this paper is related to hybrid neuro-symbolic approaches for inference with natural language (Liu et al., 2020; Minervini et al., 2020; Jiang and Bansal, 2019; Chen et al., 2019; Dua et al., 2019; Xu et al., 2021; Weber et al., 2019).”