@inproceedings{xiao-etal-2026-long,
title = "Long Document Reconstruction Unlocks Scalable Long-Context {RLVR}",
author = "Xiao, Yao and
Wang, Lei and
Deng, Yue and
Chen, Guanzheng and
Jin, Ziqi and
Kim, Jung-jae and
Li, Xiaoli and
Lee, Roy Ka-Wei and
Bing, Lidong",
editor = "Lee, Hung-Yi and
Dev, Sunipa and
Li, Jessy and
Elliott, Desmond",
booktitle = "Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing",
month = oct,
year = "2026",
address = "Budapest, Hungary",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.emnlp-main.135/",
pages = "3122--3135",
ISBN = "979-8-89176-459-0",
abstract = "Reinforcement Learning with Verifiable Rewards (RLVR) has become a prominent paradigm to enhance the capabilities (i.e. long-context) of Large Language Models (LLMs). However, it often relies on gold-standard answers or explicit evaluation rubrics provided by powerful teacher models or human experts, which are costly and time-consuming. In this work, we investigate unsupervised approaches to enhance the long-context capabilities of LLMs, eliminating the need for heavy human annotations or teacher models' supervision. Specifically, we first replace a few paragraphs with special placeholders in a long document. LLMs are then trained through reinforcement learning to reconstruct the long document by correctly identifying and sequencing missing paragraphs from a set of candidate options. This training paradigm enables the model to capture global narrative coherence, significantly boosting long-context performance. We validate the effectiveness of our method on two widely used benchmarks, RULER and LongBench v2. While acquiring noticeable gains on RULER (nearly 10 points), it can also achieve a reasonable improvement on LongBench v2 without any manually curated long-context QA data. Furthermore, we conduct extensive ablation studies to analyze the impact of reward designs, data curation strategies, training schemes, and data scaling effects on model performance. We will release our code, data, and models."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="xiao-etal-2026-long">
<titleInfo>
<title>Long Document Reconstruction Unlocks Scalable Long-Context RLVR</title>
</titleInfo>
<name type="personal">
<namePart type="given">Yao</namePart>
<namePart type="family">Xiao</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Lei</namePart>
<namePart type="family">Wang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yue</namePart>
<namePart type="family">Deng</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Guanzheng</namePart>
<namePart type="family">Chen</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ziqi</namePart>
<namePart type="family">Jin</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jung-jae</namePart>
<namePart type="family">Kim</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Xiaoli</namePart>
<namePart type="family">Li</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Roy</namePart>
<namePart type="given">Ka-Wei</namePart>
<namePart type="family">Lee</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Lidong</namePart>
<namePart type="family">Bing</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2026-10</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing</title>
</titleInfo>
<name type="personal">
<namePart type="given">Hung-Yi</namePart>
<namePart type="family">Lee</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Sunipa</namePart>
<namePart type="family">Dev</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jessy</namePart>
<namePart type="family">Li</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Desmond</namePart>
<namePart type="family">Elliott</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Budapest, Hungary</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
<identifier type="isbn">979-8-89176-459-0</identifier>
</relatedItem>
<abstract>Reinforcement Learning with Verifiable Rewards (RLVR) has become a prominent paradigm to enhance the capabilities (i.e. long-context) of Large Language Models (LLMs). However, it often relies on gold-standard answers or explicit evaluation rubrics provided by powerful teacher models or human experts, which are costly and time-consuming. In this work, we investigate unsupervised approaches to enhance the long-context capabilities of LLMs, eliminating the need for heavy human annotations or teacher models’ supervision. Specifically, we first replace a few paragraphs with special placeholders in a long document. LLMs are then trained through reinforcement learning to reconstruct the long document by correctly identifying and sequencing missing paragraphs from a set of candidate options. This training paradigm enables the model to capture global narrative coherence, significantly boosting long-context performance. We validate the effectiveness of our method on two widely used benchmarks, RULER and LongBench v2. While acquiring noticeable gains on RULER (nearly 10 points), it can also achieve a reasonable improvement on LongBench v2 without any manually curated long-context QA data. Furthermore, we conduct extensive ablation studies to analyze the impact of reward designs, data curation strategies, training schemes, and data scaling effects on model performance. We will release our code, data, and models.</abstract>
<identifier type="citekey">xiao-etal-2026-long</identifier>
<location>
<url>https://aclanthology.org/2026.emnlp-main.135/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>3122</start>
<end>3135</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Long Document Reconstruction Unlocks Scalable Long-Context RLVR
%A Xiao, Yao
%A Wang, Lei
%A Deng, Yue
%A Chen, Guanzheng
%A Jin, Ziqi
%A Kim, Jung-jae
%A Li, Xiaoli
%A Lee, Roy Ka-Wei
%A Bing, Lidong
%Y Lee, Hung-Yi
%Y Dev, Sunipa
%Y Li, Jessy
%Y Elliott, Desmond
%S Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing
%D 2026
%8 October
%I Association for Computational Linguistics
%C Budapest, Hungary
%@ 979-8-89176-459-0
%F xiao-etal-2026-long
%X Reinforcement Learning with Verifiable Rewards (RLVR) has become a prominent paradigm to enhance the capabilities (i.e. long-context) of Large Language Models (LLMs). However, it often relies on gold-standard answers or explicit evaluation rubrics provided by powerful teacher models or human experts, which are costly and time-consuming. In this work, we investigate unsupervised approaches to enhance the long-context capabilities of LLMs, eliminating the need for heavy human annotations or teacher models’ supervision. Specifically, we first replace a few paragraphs with special placeholders in a long document. LLMs are then trained through reinforcement learning to reconstruct the long document by correctly identifying and sequencing missing paragraphs from a set of candidate options. This training paradigm enables the model to capture global narrative coherence, significantly boosting long-context performance. We validate the effectiveness of our method on two widely used benchmarks, RULER and LongBench v2. While acquiring noticeable gains on RULER (nearly 10 points), it can also achieve a reasonable improvement on LongBench v2 without any manually curated long-context QA data. Furthermore, we conduct extensive ablation studies to analyze the impact of reward designs, data curation strategies, training schemes, and data scaling effects on model performance. We will release our code, data, and models.
%U https://aclanthology.org/2026.emnlp-main.135/
%P 3122-3135
Markdown (Informal)
[Long Document Reconstruction Unlocks Scalable Long-Context RLVR](https://aclanthology.org/2026.emnlp-main.135/) (Xiao et al., EMNLP 2026)
ACL
- Yao Xiao, Lei Wang, Yue Deng, Guanzheng Chen, Ziqi Jin, Jung-jae Kim, Xiaoli Li, Roy Ka-Wei Lee, and Lidong Bing. 2026. Long Document Reconstruction Unlocks Scalable Long-Context RLVR. In Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing, pages 3122–3135, Budapest, Hungary. Association for Computational Linguistics.