@inproceedings{upreti-2026-plan,
title = "Where Does Plan Validity Live in Grammar-Constrained {LLM} Planning? An Applicability Mask Induced from Environment Rollouts",
author = "Upreti, Nijesh",
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.150/",
pages = "2965--2990",
ISBN = "979-8-89176-444-6",
abstract = "LLM-augmented symbolic planning combines natural-language goal grounding with symbolic search and constrained decoding. We present Abstraction-based Grammar-Constrained Planning (AGCP), a framework that induces lifted action schemas from environment rollouts, compiles a bounded language of grounded action sequences that are applicable and goal-reaching under the induced model, and enforces that language during decoding. The guarantee is relative to the induced model. Environment execution additionally requires a sound induced model and a correct parsed goal. With the model, prompts, instances, and decoding settings fixed, replacing a syntax-only grammar with the applicability mask, compiled here from the domain{'}s own transition model, moves goal-reaching samples from 0/160 to 160/160. On human-annotated ALFWorld goals, verifier-guided correction raises parse acceptance from 61.4{\%} to 75.4{\%} and execution success from 67.5{\%} to 80.7{\%} on 114 paired tasks. Under paraphrasing, GPT-4o and Sonnet grounding remain stable while a regular-expression baseline falls from 31.9{\%} to 0.8{\%}. Together, these results locate validity relative to the induced model in an applicability grammar built from a symbolically observable, bounded environment. In this setting, natural-language goal grounding is the main remaining role of the LLM."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="upreti-2026-plan">
<titleInfo>
<title>Where Does Plan Validity Live in Grammar-Constrained LLM Planning? An Applicability Mask Induced from Environment Rollouts</title>
</titleInfo>
<name type="personal">
<namePart type="given">Nijesh</namePart>
<namePart type="family">Upreti</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>LLM-augmented symbolic planning combines natural-language goal grounding with symbolic search and constrained decoding. We present Abstraction-based Grammar-Constrained Planning (AGCP), a framework that induces lifted action schemas from environment rollouts, compiles a bounded language of grounded action sequences that are applicable and goal-reaching under the induced model, and enforces that language during decoding. The guarantee is relative to the induced model. Environment execution additionally requires a sound induced model and a correct parsed goal. With the model, prompts, instances, and decoding settings fixed, replacing a syntax-only grammar with the applicability mask, compiled here from the domain’s own transition model, moves goal-reaching samples from 0/160 to 160/160. On human-annotated ALFWorld goals, verifier-guided correction raises parse acceptance from 61.4% to 75.4% and execution success from 67.5% to 80.7% on 114 paired tasks. Under paraphrasing, GPT-4o and Sonnet grounding remain stable while a regular-expression baseline falls from 31.9% to 0.8%. Together, these results locate validity relative to the induced model in an applicability grammar built from a symbolically observable, bounded environment. In this setting, natural-language goal grounding is the main remaining role of the LLM.</abstract>
<identifier type="citekey">upreti-2026-plan</identifier>
<location>
<url>https://aclanthology.org/2026.findings-aacl.150/</url>
</location>
<part>
<date>2026-11</date>
<extent unit="page">
<start>2965</start>
<end>2990</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Where Does Plan Validity Live in Grammar-Constrained LLM Planning? An Applicability Mask Induced from Environment Rollouts
%A Upreti, Nijesh
%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 upreti-2026-plan
%X LLM-augmented symbolic planning combines natural-language goal grounding with symbolic search and constrained decoding. We present Abstraction-based Grammar-Constrained Planning (AGCP), a framework that induces lifted action schemas from environment rollouts, compiles a bounded language of grounded action sequences that are applicable and goal-reaching under the induced model, and enforces that language during decoding. The guarantee is relative to the induced model. Environment execution additionally requires a sound induced model and a correct parsed goal. With the model, prompts, instances, and decoding settings fixed, replacing a syntax-only grammar with the applicability mask, compiled here from the domain’s own transition model, moves goal-reaching samples from 0/160 to 160/160. On human-annotated ALFWorld goals, verifier-guided correction raises parse acceptance from 61.4% to 75.4% and execution success from 67.5% to 80.7% on 114 paired tasks. Under paraphrasing, GPT-4o and Sonnet grounding remain stable while a regular-expression baseline falls from 31.9% to 0.8%. Together, these results locate validity relative to the induced model in an applicability grammar built from a symbolically observable, bounded environment. In this setting, natural-language goal grounding is the main remaining role of the LLM.
%U https://aclanthology.org/2026.findings-aacl.150/
%P 2965-2990
Markdown (Informal)
[Where Does Plan Validity Live in Grammar-Constrained LLM Planning? An Applicability Mask Induced from Environment Rollouts](https://aclanthology.org/2026.findings-aacl.150/) (Upreti, Findings 2026)
ACL