Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

ReCo: Reliable Causal Chain Reasoning via Structural Causal Recurrent Neural Networks

Kai Xiong, Xiao Ding, Zhongyang Li, Li Du, Bing Qin, Yi Zheng, Baoxing Huai

TL;DR

By injecting reliable causal chain knowledge distilled by ReCo, BERT can achieve better performances on four downstream causal-related tasks than BERT models enhanced by other kinds of knowledge.

Abstract

Causal chain reasoning (CCR) is an essential ability for many decision-making AI systems, which requires the model to build reliable causal chains by connecting causal pairs. However, CCR suffers from two main transitive problems: threshold effect and scene drift. In other words, the causal pairs to be spliced may have a conflicting threshold boundary or scenario. To address these issues, we propose a novel Reliable Causal chain reasoning framework~(ReCo), which introduces exogenous variables to represent the threshold and scene factors of each causal pair within the causal chain, and estimates the threshold and scene contradictions across exogenous variables via structural causal recurrent neural networks~(SRNN). Experiments show that ReCo outperforms a series of strong baselines on both Chinese and English CCR datasets. Moreover, by injecting reliable causal chain knowledge distilled by ReCo, BERT can achieve better performances on four downstream causal-related tasks than BERT models enhanced by other kinds of knowledge.

ReCo: Reliable Causal Chain Reasoning via Structural Causal Recurrent Neural Networks

TL;DR

By injecting reliable causal chain knowledge distilled by ReCo, BERT can achieve better performances on four downstream causal-related tasks than BERT models enhanced by other kinds of knowledge.

Abstract

Causal chain reasoning (CCR) is an essential ability for many decision-making AI systems, which requires the model to build reliable causal chains by connecting causal pairs. However, CCR suffers from two main transitive problems: threshold effect and scene drift. In other words, the causal pairs to be spliced may have a conflicting threshold boundary or scenario. To address these issues, we propose a novel Reliable Causal chain reasoning framework~(ReCo), which introduces exogenous variables to represent the threshold and scene factors of each causal pair within the causal chain, and estimates the threshold and scene contradictions across exogenous variables via structural causal recurrent neural networks~(SRNN). Experiments show that ReCo outperforms a series of strong baselines on both Chinese and English CCR datasets. Moreover, by injecting reliable causal chain knowledge distilled by ReCo, BERT can achieve better performances on four downstream causal-related tasks than BERT models enhanced by other kinds of knowledge.
Paper Structure (48 sections, 11 equations, 4 figures, 10 tables)

This paper contains 48 sections, 11 equations, 4 figures, 10 tables.

Table of Contents

  1. Introduction
  2. Background
  3. Problem Definition
  4. Structural Causal Model
  5. Method
  6. Overview
  7. Encoder
  8. Exogenous-aware CVAE
  9. Structural Causal Recurrent Neural Networks
  10. Scene Drift Estimator
  11. Hidden Gate
  12. Threshold Effect Estimator
  13. Exogenous Gate
  14. Output Gate
  15. Problems and Reliability Predictor
  16. ...and 33 more sections

Figures (4)

  • Figure 1: Causal chains with (a) threshold effect and (b) scene drift problems, which can be estimated by the contradictions of threshold and scene factors in the contexts, respectively.
  • Figure 2: (a) Constructing SCM based on an antecedent causal chain and a causal pair. (b) If there is threshold effect or scene drift problem, then $U_{xy}$ would contradict $U_{yx}$. And it is worth discussing the threshold effect problem when scenes are consistent.
  • Figure 3: (a) The overall architecture of ReCo. (b) The detailed structure of EA-CVAE. (c) The detailed structure of SRNN which is a kind of recurrent neural networks.
  • Figure 4: Accuracy on (a) Chinese and (b) English CCR test sets categorized by the lengths of the causal chains.