PonsRAG: A Pons-Inspired RAG Bridging Cognitive Islands for Coordinated Long Narrative Reasoning

Rongchen Zhao; Yu Chen; Juyuan Wang; Zhouting Mo; Jianxing Yu; Wenqing Chen; Jingping Liu

cs.AIAug 26, 2026

PonsRAG: A Pons-Inspired RAG Bridging Cognitive Islands for Coordinated Long Narrative Reasoning

Authors: Rongchen Zhao, Yu Chen, Juyuan Wang, Zhouting Mo, Jianxing Yu, Wenqing Chen, Jingping Liu

Organizations: School of Software Engineering, Sun Yat-sen University · School of Future Technology, South China University of Technology · TikTok Inc · School of Artificial Intelligence, Sun Yat-sen University

Abstract

Long Narrative Reasoning is an essential capability for processing and reasoning over complex narratives. While retrieval-augmented generation provides a promising framework, existing methods still face two critical challenges: cognitive islanding and cross-layer evidence disconnection. To address these issues, we propose PonsRAG, a coordinated RAG framework inspired by the biological pons. PonsRAG consists of two key components: Triple-Layer Indexing, which organizes documents into a connected knowledge structure to bridge cognitive islands, and Coordinated Reasoning, which retrieves evidence across distinct layers and integrates cross-layer information into a unified context. We evaluate PonsRAG on four long-context narrative benchmarks, and experimental results show that it outperforms the strongest baseline, achieving a 11.56% relative improvement in average accuracy on multi-choice tasks.

Figures & tables

Figure 1: Bridging cognitive islands in long narrative reasoning.

Figure 2: Overall architecture of PonsRAG.

Method	NarrativeQA		EN.QA		EN.MC	NoCha	QA Avg.		MC Avg.
	F1	EM	F1	EM	ACC	ACC	F1	EM	ACC
LLM
GPT-4o-mini	27.29	7.00	29.83	12.82	30.57	60.32	28.56	9.91	45.45
Naive RAG
BGE-M3(0.3B)	23.16	15.10	23.71	16.24	59.82	56.35	23.44	15.67	58.09
NV-Embed-v2 (7B)	27.18	17.80	34.34	24.57	61.13	68.25	30.76	21.19	64.69

Table 1: Single-step QA performance on four long narrative comprehension datasets. For fair comparison, we adopt GPT-4o-mini as the LLM backbone. For one-step setting, we limit ComoRAG for max one step. We highlight the best and second-best results. Improv. denotes the relative improvement of our method over the second-best baseline.

Method	NarrativeQA		EN.QA		EN.MC	NoCha	QA Avg.		MC Avg.
	F1	EM	F1	EM	ACC	ACC	F1	EM	ACC
IRCoT +RAPTOR	31.35	16.00	32.09	19.36	63.76	57.94	31.72	17.68	60.85
IRCoT +HippoRAGv2	28.98	13.00	29.27	18.24	64.19	61.90	29.13	15.62	63.05
IRCoT +Youtu-GraphRAG	25.92	15.40	31.10	22.79	66.38	63.49	28.51	19.10	64.94
ComoRAG (five steps)	31.43	18.60	34.52	25.07	72.93	61.90	32.98	21.84	67.42
IRCoT+PonsRAG (Ours)	33.28	20.60	36.30	26.78	79.48	72.22	34.79	23.69	75.85

Table 2: Multi-step QA performance on four long-narrative comprehension datasets. For fair comparison, we use IRCoT for all enhanced RAGs except ComoRAG, which follows its original multi-step setting with max five steps (details in Section 5.1 ).

Figure 3: Performance by query types of RAG methods.

Index
Method	EN.MC	EN.QA
	ACC	F1	EM
PonsRAG	77.73	35.13	26.21
w/ Char	52.40	21.73	17.95
w/ Plot	55.02	23.37	18.23
w/o Pons	61.13	28.59	19.37

Table 3: Ablation studies of PonsRAG.

Panel of Figure 4 — Figure 4: Analysis of island density and performance gains.

$\tau$	EN.MC (ACC)	EN.QA (F1)
0.00	$72.73\pm 0.51$	$31.75\pm 0.59$
0.25	$73.62\pm 0.34$	$32.58\pm 0.41$
0.50	$77.31\pm 0.21$	$\textbf{34.61}\pm 0.34$
0.75	$\textbf{77.60}\pm\textbf{0.11}$	$34.03\pm\textbf{0.20}$
0.80	$74.63\pm 0.12$	$33.12\pm 0.17$
0.90	$74.20\pm 0.09$	$32.35\pm 0.13$

Table 4: Performance under diverse Sparsity Controller

\tau

(details in Equation ( 4 )).

Matching Strategy	EN.MC (ACC)	EN.QA (F1)	Avg Context Tokens
Dense (w/o Match)	68.12	28.65	5,203
1-to-3	73.03	30.34	3,856
1-to-2	74.24	32.28	2,438
1-to-1 (Ours)	77.73	35.13	1,187

Table 5: Ablation on maximum degree constraints in Pons Match.

Appendix figures & tables8 assets

Supplementary material from the paper’s appendix.

Appendix

Dataset	#Docs	#Queries	Total Tokens	Avg. Tokens
NarrativeQA	17	500	887,763	52,221
EN.QA	69	351	14,497,149	210,104
EN.MC	58	229	11,249,572	193,958
NoCha	4	126	555,972	138,993

Appendix

Table 6: Statistics of benchmarks. Avg. Tokens denotes the average length of documents.

Model Selection
Hyperparameter	Value
LLM Backbone Agent	GPT-4o-mini
Retrieval Embedding Model	BGE-M3
Triple-Layer Indexing Setup
Max Chunk Size	512 tokens
Sparsity Controller ( $\tau$ )	0.75 (MC) / 0.50 (QA)

Appendix

Table 7: Detailed hyperparameter settings of PonsRAG.

MGS Configuration	EN.MC (ACC)	EN.QA (F1)
$\alpha=1.0,\beta=0.0,\gamma=0.0$	$74.02\pm 0.42$	$33.95\pm 0.55$
$\alpha=0.0,\beta=1.0,\gamma=0.0$	$69.53\pm 0.48$	$32.08\pm 0.49$
$\alpha=0.0,\beta=0.0,\gamma=1.0$	$67.74\pm 0.39$	$31.15\pm 0.51$
$\alpha=0.5,\beta=0.1,\gamma=0.4$	$74.25\pm 0.65$	$33.36\pm 0.71$
$\alpha=0.5,\beta=0.2,\gamma=0.3$	$74.89\pm 0.48$	$34.01\pm 0.54$
$\alpha=0.6,\beta=0.1,\gamma=0.3$	$75.78\pm 0.27$	$34.67\pm 0.32$

Appendix

Table 8: Performance of PonsRAG with different MGS configurations at

\tau=0.75

for EN.MC and

\tau=0.50

for EN.QA.

Bridging Scheme	Prec@100	EN.MC	EN.QA
		(ACC)	(F1)
Entity Mention	91.0%	58.45	24.82
Pure Semantic	64.0%	68.12	30.95
Pons Weight (Ours)	82.0%	75.54	34.56

Appendix

Table 9: Manual precision of edges and downstream performance on EN.MC and EN.QA across different bridging schemes. Downstream results for Ours are consistent with Table 3 .

Query Type	EN.MC	EN.QA	Total	Proportion
Char	59	114	173	29.8%
Plot	68	86	154	26.6%
Mix	102	151	253	43.6%
Total	229	351	580	100.0%

Appendix

Table 10: Detailed distribution and counts of query types across the EN.MC and EN.QA datasets.

	Predicted (GPT-4o)
True (Human)	Char	Plot	Mix	Total
Char	25	3	1	29
Plot	1	22	3	26
Mix	0	4	41	45
Accuracy	88.0%

Appendix

Table 11: Confusion matrix of LLM-based query classification against human annotation on 100 sampled queries.

Performance
Metrics	PonsRAG	GraphRAG	HiRAG
F1 Score	34.38 (100%)	14.60 (42.5%)	21.37 (62.2%)
EM Score	25.31 (100%)	8.20 (32.4%)	14.28 (56.4%)
Token Usage
Tokens	1.08M (100%)	26.43M (2447%)	95.81M (8871%)
Average Time (s)

Appendix

Table 12: Comparison of performance, token usage, and latency across different RAG paradigms. Percentages indicate the relative ratio compared to PonsRAG.

Input Data (No Options)

Query: How does Jimmy Doyle spend all of his money with his friends in “After the Race”?

Options: [A] Betting on car racing [B] Playing cards [C] Arm wrestling [D] Treating his friends to drinks

PonsRAG’s Choice Result

Query Anchor

Anchor Char Nodes

\mathcal{V}_{anc}^{char}

- JIMMY (PERSON) : Jimmy is the son of a wealthy butcher, educated in England and known for his popularity and social life…

Appendix

Table 13: Case Study on Coordinated Narrative Reasoning. We present a case to demonstrate our model’s performance in long-context understanding. Different colors are used to highlight the nature of the processed information: Green is used for key cognition found in step Query Anchor that contributes to the correct answer, while Purple is used for the evidence in step Pons Awaken .

Retrieval-Augmented Generation

Retrieval-augmented generation connects language models to external knowledge so answers can use current, domain-specific evidence.

See all 792 papers

Hit-RAG: Learning to Reason with Long Contexts via Preference Alignment
Mar 7, 2026Junming Liu, Yuqi Li, Shiping Wen +2Retrieval-Augmented Generation Multimodal RAG
Tongji University, Shanghai, China · The City University of New York, New York, USA · University of Technology Sydney, Sydney, Australia +2
ClueWeaver: Reward-Guided Dual-Agent Evidence Reasoning for Compact LLMs on Literary Long Narratives
Aug 26, 2026Jihao Zhu, Zhiwei Yang, Wenxiao Zhang +7Long-Context QA Multi-Agent LLM Systems
University of Aberdeen, Aberdeen, United Kingdom · Institute of Information Engineering, CAS, Beijing, China · The University of Western Australia, Perth, Australia +1
CodaRAG: Connecting the Dots with Associativity Inspired by Complementary Learning
Apr 12, 2026Cheng-Yen Li, Xuanjun Chen, Claire Lin +4Retrieval-Augmented Generation Graph Retrieval-Augmented Generation
National Taiwan University, Taipei, Taiwan · NTU Artificial Intelligence Center of Research Excellence, National Taiwan University.

Mar 7, 2026J/K move · Enter open · S save

Hit-RAG: Learning to Reason with Long Contexts via Preference Alignment

Junming Liu, Yuqi Li, Shiping Wen, Zhigang Zeng+1

Tongji University, Shanghai, China · The City University of New York, New York, USA · University of Technology Sydney, Sydney, Australia+2

Retrieval-Augmented Generation Multimodal RAG LLM Alignment Multimodal Large Language Models

Open paper AskPDF arXiv

Despite the promise of Retrieval-Augmented Generation in grounding Multimodal Large Language Models with external knowledge, the transition to extensive contexts often leads to significant attention dilution and reasoning hallucinations. The surge in information density causes critical evidence to be submerged by voluminous noise, which complicates the discernment of relevant fragments within a dense input. In this paper, we propose Hit-RAG, a multi-stage preference alignment framework designed to resolve these cognitive bottlenecks through a progressive optimization pipeline. Our approach systematically refines the utilization of external evidence via three distinct stages. First, Supervised Fine-tuning establishes baseline context awareness to minimize information neglect. Next, Discriminative Preference Alignment enhances robustness against misleading distractors. Finally, Group-Relative Policy Optimization stabilizes logical synthesis to prevent reasoning collapse. Extensive evaluations on eight benchmarks demonstrate that Hit-RAG consistently yields substantial performance gains, enabling models to bridge the gap between context acquisition and accurate reasoning while surpassing much larger counterparts in long-context scenarios.

Abstract

Retrieval-Augmented Generation

Explore similar work

Hit-RAG: Learning to Reason with Long Contexts via Preference Alignment

ClueWeaver: Reward-Guided Dual-Agent Evidence Reasoning for Compact LLMs on Literary Long Narratives

CodaRAG: Connecting the Dots with Associativity Inspired by Complementary Learning