@inproceedings{yin-etal-2026-learning,
title = "Learning Composable Chains-of-Thought",
author = "Yin, Fangcong and
Liu, Zeyu Leo and
Leqi, Liu and
Ye, Xi and
Durrett, Greg",
editor = "Lee, Hung-Yi and
Dev, Sunipa and
Li, Jessy and
Elliott, Desmond",
booktitle = "Findings of the {A}ssociation for {C}omputational {L}inguistics: {EMNLP} 2026",
month = oct,
year = "2026",
address = "Budapest, Hungary",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.findings-emnlp.1408/",
doi = "10.18653/2026.findings-emnlp.1408",
pages = "30018--30044",
ISBN = "979-8-89176-462-0",
abstract = "A common approach for teaching large language models (LLMs) to reason is to train on chains-of-thought (CoTs) of in-distribution reasoning problems. However, such annotated data is costly to obtain for every problem of interest. We want reasoning models to generalize beyond their training distribution, and ideally to generalize compositionally: they should combine atomic reasoning skills to solve harder unseen tasks. In this paper, we introduce a method to enable generalization to a target compositional task that has no labeled CoT data. We minimally modify CoT formats of constituent atomic tasks to be composable: we add prefixes to CoTs during training time, thereby encouraging the trained model to generate sequences of CoTs during inference. We train individual models on the atomic tasks with composable CoT data and combine them with multitask learning or model merging to address the target compositional task zero-shot. This model can be further trained on a small amount of compositional data using rejection sampling fine-tuning (RFT). Training LLMs on Composable CoT outperforms multitask learning and continued fine-tuning baselines on three domains of controlled compositional tasks and also yields improvement on math skill composition setting derived from the MATH dataset."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="yin-etal-2026-learning">
<titleInfo>
<title>Learning Composable Chains-of-Thought</title>
</titleInfo>
<name type="personal">
<namePart type="given">Fangcong</namePart>
<namePart type="family">Yin</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Zeyu</namePart>
<namePart type="given">Leo</namePart>
<namePart type="family">Liu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Liu</namePart>
<namePart type="family">Leqi</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Xi</namePart>
<namePart type="family">Ye</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Greg</namePart>
<namePart type="family">Durrett</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>Findings of the Association for Computational Linguistics: EMNLP 2026</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">979-8-89176-462-0</identifier>
</relatedItem>
<abstract>A common approach for teaching large language models (LLMs) to reason is to train on chains-of-thought (CoTs) of in-distribution reasoning problems. However, such annotated data is costly to obtain for every problem of interest. We want reasoning models to generalize beyond their training distribution, and ideally to generalize compositionally: they should combine atomic reasoning skills to solve harder unseen tasks. In this paper, we introduce a method to enable generalization to a target compositional task that has no labeled CoT data. We minimally modify CoT formats of constituent atomic tasks to be composable: we add prefixes to CoTs during training time, thereby encouraging the trained model to generate sequences of CoTs during inference. We train individual models on the atomic tasks with composable CoT data and combine them with multitask learning or model merging to address the target compositional task zero-shot. This model can be further trained on a small amount of compositional data using rejection sampling fine-tuning (RFT). Training LLMs on Composable CoT outperforms multitask learning and continued fine-tuning baselines on three domains of controlled compositional tasks and also yields improvement on math skill composition setting derived from the MATH dataset.</abstract>
<identifier type="citekey">yin-etal-2026-learning</identifier>
<identifier type="doi">10.18653/2026.findings-emnlp.1408</identifier>
<location>
<url>https://aclanthology.org/2026.findings-emnlp.1408/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>30018</start>
<end>30044</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Learning Composable Chains-of-Thought
%A Yin, Fangcong
%A Liu, Zeyu Leo
%A Leqi, Liu
%A Ye, Xi
%A Durrett, Greg
%Y Lee, Hung-Yi
%Y Dev, Sunipa
%Y Li, Jessy
%Y Elliott, Desmond
%S Findings of the Association for Computational Linguistics: EMNLP 2026
%D 2026
%8 October
%I Association for Computational Linguistics
%C Budapest, Hungary
%@ 979-8-89176-462-0
%F yin-etal-2026-learning
%X A common approach for teaching large language models (LLMs) to reason is to train on chains-of-thought (CoTs) of in-distribution reasoning problems. However, such annotated data is costly to obtain for every problem of interest. We want reasoning models to generalize beyond their training distribution, and ideally to generalize compositionally: they should combine atomic reasoning skills to solve harder unseen tasks. In this paper, we introduce a method to enable generalization to a target compositional task that has no labeled CoT data. We minimally modify CoT formats of constituent atomic tasks to be composable: we add prefixes to CoTs during training time, thereby encouraging the trained model to generate sequences of CoTs during inference. We train individual models on the atomic tasks with composable CoT data and combine them with multitask learning or model merging to address the target compositional task zero-shot. This model can be further trained on a small amount of compositional data using rejection sampling fine-tuning (RFT). Training LLMs on Composable CoT outperforms multitask learning and continued fine-tuning baselines on three domains of controlled compositional tasks and also yields improvement on math skill composition setting derived from the MATH dataset.
%R 10.18653/2026.findings-emnlp.1408
%U https://aclanthology.org/2026.findings-emnlp.1408/
%U https://doi.org/10.18653/2026.findings-emnlp.1408
%P 30018-30044
Markdown (Informal)
[Learning Composable Chains-of-Thought](https://aclanthology.org/2026.findings-emnlp.1408/) (Yin et al., Findings 2026)
ACL
- Fangcong Yin, Zeyu Leo Liu, Liu Leqi, Xi Ye, and Greg Durrett. 2026. Learning Composable Chains-of-Thought. In Findings of the Association for Computational Linguistics: EMNLP 2026, pages 30018–30044, Budapest, Hungary. Association for Computational Linguistics.