@inproceedings{nguyen-etal-2026-copy,
title = "Copy Less, Ground More: Selective and Confidence-Aware Decoding for Grounded Legal Generation",
author = "Nguyen, Tan-Minh and
Do, Truong Dinh and
Nguyen, Hiep K. V. and
Nguyen, Le-Minh",
editor = "Zhang, Yue and
Xu, Ruifeng and
Purwarianti, Ayu and
Rong, Tong and
Suzuki, Jun",
booktitle = "Findings of the {A}ssociation for {C}omputational {L}inguistics: {IJCNLP}-{AACL} 2026",
month = nov,
year = "2026",
address = "Hengqin, China",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.findings-aacl.158/",
pages = "3144--3168",
ISBN = "979-8-89176-444-6",
abstract = "While retrieval-augmented generation (RAG) grounds large language models (LLMs) in external knowledge, ensuring faithful evidence utilization in complex domains like law remains challenging. LLMs often ignore retrieved facts or indiscriminately copy from noisy contexts. Existing context-aware decoding methods attempt to mitigate this by amplifying the influence of retrieved content, but they treat the context as a monolithic block{---}amplifying irrelevant spans alongside evidence{---}and intervene at every decoding step, incurring substantial inference overhead. To address this, we propose \textsc{SaCoD} (Selective and Confidence-aware Decoding), a training-free inference framework. \textsc{SaCoD} leverages the LLM{'}s internal attention signals to dynamically isolate salient evidence paragraphs at each step, restricting a customized copy mechanism strictly to relevant facts. Furthermore, an entropy-based gate activates this intervention only when the model is uncertain, bypassing copy operations for deterministic tokens to preserve linguistic fluency and cut computational overhead. Experiments on four legal question-answering benchmarks across three open-weight LLMs demonstrate that \textsc{SaCoD} significantly improves faithfulness and response correctness while substantially reducing inference latency compared with prior context-aware methods."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="nguyen-etal-2026-copy">
<titleInfo>
<title>Copy Less, Ground More: Selective and Confidence-Aware Decoding for Grounded Legal Generation</title>
</titleInfo>
<name type="personal">
<namePart type="given">Tan-Minh</namePart>
<namePart type="family">Nguyen</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Truong</namePart>
<namePart type="given">Dinh</namePart>
<namePart type="family">Do</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Hiep</namePart>
<namePart type="given">K</namePart>
<namePart type="given">V</namePart>
<namePart type="family">Nguyen</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Le-Minh</namePart>
<namePart type="family">Nguyen</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2026-11</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Findings of the Association for Computational Linguistics: IJCNLP-AACL 2026</title>
</titleInfo>
<name type="personal">
<namePart type="given">Yue</namePart>
<namePart type="family">Zhang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ruifeng</namePart>
<namePart type="family">Xu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ayu</namePart>
<namePart type="family">Purwarianti</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Tong</namePart>
<namePart type="family">Rong</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jun</namePart>
<namePart type="family">Suzuki</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Hengqin, China</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
<identifier type="isbn">979-8-89176-444-6</identifier>
</relatedItem>
<abstract>While retrieval-augmented generation (RAG) grounds large language models (LLMs) in external knowledge, ensuring faithful evidence utilization in complex domains like law remains challenging. LLMs often ignore retrieved facts or indiscriminately copy from noisy contexts. Existing context-aware decoding methods attempt to mitigate this by amplifying the influence of retrieved content, but they treat the context as a monolithic block—amplifying irrelevant spans alongside evidence—and intervene at every decoding step, incurring substantial inference overhead. To address this, we propose SaCoD (Selective and Confidence-aware Decoding), a training-free inference framework. SaCoD leverages the LLM’s internal attention signals to dynamically isolate salient evidence paragraphs at each step, restricting a customized copy mechanism strictly to relevant facts. Furthermore, an entropy-based gate activates this intervention only when the model is uncertain, bypassing copy operations for deterministic tokens to preserve linguistic fluency and cut computational overhead. Experiments on four legal question-answering benchmarks across three open-weight LLMs demonstrate that SaCoD significantly improves faithfulness and response correctness while substantially reducing inference latency compared with prior context-aware methods.</abstract>
<identifier type="citekey">nguyen-etal-2026-copy</identifier>
<location>
<url>https://aclanthology.org/2026.findings-aacl.158/</url>
</location>
<part>
<date>2026-11</date>
<extent unit="page">
<start>3144</start>
<end>3168</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Copy Less, Ground More: Selective and Confidence-Aware Decoding for Grounded Legal Generation
%A Nguyen, Tan-Minh
%A Do, Truong Dinh
%A Nguyen, Hiep K. V.
%A Nguyen, Le-Minh
%Y Zhang, Yue
%Y Xu, Ruifeng
%Y Purwarianti, Ayu
%Y Rong, Tong
%Y Suzuki, Jun
%S Findings of the Association for Computational Linguistics: IJCNLP-AACL 2026
%D 2026
%8 November
%I Association for Computational Linguistics
%C Hengqin, China
%@ 979-8-89176-444-6
%F nguyen-etal-2026-copy
%X While retrieval-augmented generation (RAG) grounds large language models (LLMs) in external knowledge, ensuring faithful evidence utilization in complex domains like law remains challenging. LLMs often ignore retrieved facts or indiscriminately copy from noisy contexts. Existing context-aware decoding methods attempt to mitigate this by amplifying the influence of retrieved content, but they treat the context as a monolithic block—amplifying irrelevant spans alongside evidence—and intervene at every decoding step, incurring substantial inference overhead. To address this, we propose SaCoD (Selective and Confidence-aware Decoding), a training-free inference framework. SaCoD leverages the LLM’s internal attention signals to dynamically isolate salient evidence paragraphs at each step, restricting a customized copy mechanism strictly to relevant facts. Furthermore, an entropy-based gate activates this intervention only when the model is uncertain, bypassing copy operations for deterministic tokens to preserve linguistic fluency and cut computational overhead. Experiments on four legal question-answering benchmarks across three open-weight LLMs demonstrate that SaCoD significantly improves faithfulness and response correctness while substantially reducing inference latency compared with prior context-aware methods.
%U https://aclanthology.org/2026.findings-aacl.158/
%P 3144-3168
Markdown (Informal)
[Copy Less, Ground More: Selective and Confidence-Aware Decoding for Grounded Legal Generation](https://aclanthology.org/2026.findings-aacl.158/) (Nguyen et al., Findings 2026)
ACL