Research map / Vision-language and multimodal / Multimodal LLMs
Chinese Two-part Allegorical Sayings Reading Comprehension: Exploration from Reasoning to Metaphor
Dongyu Su, Yimin Xiao, Tongguan Wang, Feiyue Xue, Junkai Li, Hui Liu, Ying Sha
Read the paperCheck your idea against 137,217 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Introduces a reading comprehension benchmark for Chinese two-part allegorical sayings (TPAS) to probe semantic rhetoric, logical reasoning, and metaphor understanding.
Problem
Existing research uses Chinese two-part allegorical sayings (TPAS) mainly to assist other semantic tasks and lacks in-depth exploration of their intrinsic mechanisms: semantic rhetoric, logical reasoning, and metaphorical expression.
Approach
Constructs the first Chinese TPAS reading comprehension dataset to study the riddle-explanation structure of TPAS as a testbed for reasoning and metaphor understanding.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add joint vector cross-projection (ablation) | MTCLN · CTRC (average over five sets) | Avg-Acc (%): 66.96 -> 67.47 | Improved |
| add softened contrastive learning (ablation) | MTCLN · CTRC (average over five sets) | Avg-Acc (%): 59.18 -> 67.47 | Improved |
| add multi-view feature fusion (ablation) | MTCLN · CTRC (average over five sets) | Avg-Acc (%): 66.78 -> 67.47 | Improved |
| add softened contrastive learning on top of JP+MF (ablation) | MTCLN · CTRC (average over five sets) | Avg-Acc (%): 54.66 -> 67.47 | Improved |
| add multi-view fusion and softened contrastive learning together (no JP) (ablation) | MTCLN · CTRC (average over five sets) | Avg-Acc (%): 58.82 -> 67.47 | Improved |
| add all three modules to RoBERTa baseline (ablation) | MTCLN · CTRC (average over five sets) | Avg-Acc (%): 52.70 -> 67.47 | Improved |
| add softened contrastive learning (ablation) | MTCLN · CTRC Sim set | accuracy (%): 46.51 -> 58.34 | Improved |
| add joint vector cross-projection (ablation) | MTCLN · CTRC test | accuracy (%): 72.89 -> 73.13 | Improved |
| few-shot prompting of GPT-4o with more examples | GPT-4o-20241120 · CTRC test | accuracy (%): 42.94 -> 43.52 -> 46.32 -> 44.85 | Mixed |
| use MTCLN instead of LoRA-tuned DeepSeek-R1-1.5B / GPT-4o | MTCLN · CTRC (average over five sets) | Avg-Acc (%): outperforming LLMs by 25.48% | Improved |
Closest papers by meaning
- Mitigating Idiom Inconsistency: A Multi-Semantic Contrastive Learning Method for Chinese Idiom Reading Comprehension (AAAI 2024 · similarity 0.86)
- McHirc: A Multimodal Benchmark for Chinese Idiom Reading Comprehension (AAAI 2025 · similarity 0.86)
- BMMR: A Large-Scale Bilingual Multimodal Multi-Discipline Reasoning Dataset (NeurIPS 2025 · similarity 0.83)
- MME-SCI: A Comprehensive and Challenging Science Benchmark for Multimodal Large Language Models (AAAI 2026 · similarity 0.81)
- Thought-Path Contrastive Learning via Premise-Oriented Data Augmentation for Logical Reading Comprehension (AAAI 2025 · similarity 0.81)
- CK12: A Rounded K12 Knowledge Graph Based Benchmark for Chinese Holistic Cognition Evaluation (AAAI 2024 · similarity 0.80)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
