ThinkStruct: RST-Aware Attention for Logical Reasoning in Machine Reading Comprehension

2026-06-25 · Source: Paper Index on ACL Anthology · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Natural Language Processing · Depth: Expert, quick

Summary

ThinkStruct is a novel method addressing challenges in Machine Reading Comprehension (MRC) tasks requiring logical reasoning. Previous approaches struggled with accurate logical unit division and consistent predictions for equivalent semantics. ThinkStruct employs a transformer network enhanced with Rhetorical Structure (RS) relations. It utilizes Rhetorical Structure Theory (RST) to segment natural language text into Elementary Discourse Units (EDUs) and map their interrelationships. Node information, augmented with these logical relationships via an adjacency matrix, is fed into the transformer. Features are then integrated for answer prediction, and a contrastive learning module further refines EDU relationship understanding. Experiments show ThinkStruct outperforms state-of-the-art models on the LogiQA and Reclor datasets.

Key takeaway

For NLP engineers developing advanced Machine Reading Comprehension systems, consider integrating explicit discourse structure. ThinkStruct demonstrates that enhancing transformer networks with Rhetorical Structure Theory and contrastive learning significantly improves logical reasoning accuracy. You should explore using RST to model Elementary Discourse Units and their relationships via adjacency matrices to overcome challenges in logical unit division and prediction consistency.

Key insights

ThinkStruct improves logical reasoning in MRC by integrating Rhetorical Structure Theory into a transformer with contrastive learning.

Principles

• Rhetorical Structure improves logical reasoning.
• Explicitly modeling discourse units enhances MRC.
• Contrastive learning refines unit relationship understanding.

Method

ThinkStruct uses RST to split text into EDUs, identifies relationships, feeds node info and adjacency matrix into a transformer, integrates features, and applies contrastive learning for answer prediction.

In practice

• Apply RST for discourse unit segmentation.
• Incorporate adjacency matrices for logical relations.
• Use contrastive learning for EDU relationship refinement.

Topics

• Machine Reading Comprehension
• Logical Reasoning
• Rhetorical Structure Theory
• Transformers
• Contrastive Learning
• Elementary Discourse Units

Best for: Research Scientist, AI Scientist, Machine Learning Engineer, NLP Engineer

Related on AIssential

• Disco-RAG: Discourse-Aware Retrieval-Augmented Generation — Explicitly modeling discourse structure and planning significantly improves RAG system performance, coherence, and factual consistency.
• Structure-Aware Modeling of Multiple-Choice Questions Improves Automatic Difficulty Estimation — Explicitly modeling MCQ distractor structure significantly improves automatic difficulty estimation.
• How Do Answer Tokens Read Reasoning Traces? Self-Reading Patterns in Thinking LLMs for Quantitative Reasoning — Benign self-reading patterns in LLMs, marked by focused, forward-shifting attention, correlate with correct quantitative reasoning.
• Domain-Specialized Tree of Thought through Plug-and-Play Predictors — DST enhances LLM tree-based reasoning by using a lightweight, adaptive predictor for efficient, context-aware branch pruning.
• Decoder only Transformer : Building a GPT-2 model prototype to make it understand Natural Language… — Training a GPT-2 prototype on classic novels demonstrates decoder-only Transformer architecture and performance benefits of pre-trained weights.
• LoRi: Low-Rank Distillation for Implicit Reasoning — Hidden-state reasoning trajectories exhibit low-rank structure, enabling efficient distillation for implicit reasoning.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Paper Index on ACL Anthology.