@inproceedings{narumanchi-etal-2026-cascade,
title = "{CASCADE}: Cognitive Assessment via Sentence Context and Distortion Extraction",
author = "Narumanchi, Archit and
Rashidov, Azamat and
Park, Jong C.",
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.169/",
pages = "3376--3391",
ISBN = "979-8-89176-444-6",
abstract = "Cognitive distortions are unhelpful thinking patterns central to Cognitive Behavioral Therapy (CBT) and closely linked to mental health outcomes. Despite their context-dependent nature, studies at the intersection of Natural Language Processing and Cognitive Behavioral Therapy are centered largely around sentence-level utterances, ignoring the document-level complexity of real clinical text. To this end, we propose a two-stage rank-then-classify cascade that decomposes document-level multi-label cognitive distortion detection into (1) a sentence-ranking stage that identifies the most likely distorted sentences, and (2) a multi-label classification stage that assigns distortion subtypes to the selected sentences by incorporating their surrounding interleaved context. While the pipeline stages are trained on separate out-of-domain datasets, our model matches state-of-the-art performance on the \textsc{CBT-CD} benchmark of 166 expert-annotated clinical vignettes. Operating with only 368M parameters, our system achieves a 51.8{\%} weighted F1 (9-class), surpassing massive large language model baselines such as Mistral-7B-v0.3 (46.0{\%}) and performing on par with state-of-the-art models such as Gemma 4 26B A4B (50.8{\%})."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="narumanchi-etal-2026-cascade">
<titleInfo>
<title>CASCADE: Cognitive Assessment via Sentence Context and Distortion Extraction</title>
</titleInfo>
<name type="personal">
<namePart type="given">Archit</namePart>
<namePart type="family">Narumanchi</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Azamat</namePart>
<namePart type="family">Rashidov</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jong</namePart>
<namePart type="given">C</namePart>
<namePart type="family">Park</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>Cognitive distortions are unhelpful thinking patterns central to Cognitive Behavioral Therapy (CBT) and closely linked to mental health outcomes. Despite their context-dependent nature, studies at the intersection of Natural Language Processing and Cognitive Behavioral Therapy are centered largely around sentence-level utterances, ignoring the document-level complexity of real clinical text. To this end, we propose a two-stage rank-then-classify cascade that decomposes document-level multi-label cognitive distortion detection into (1) a sentence-ranking stage that identifies the most likely distorted sentences, and (2) a multi-label classification stage that assigns distortion subtypes to the selected sentences by incorporating their surrounding interleaved context. While the pipeline stages are trained on separate out-of-domain datasets, our model matches state-of-the-art performance on the CBT-CD benchmark of 166 expert-annotated clinical vignettes. Operating with only 368M parameters, our system achieves a 51.8% weighted F1 (9-class), surpassing massive large language model baselines such as Mistral-7B-v0.3 (46.0%) and performing on par with state-of-the-art models such as Gemma 4 26B A4B (50.8%).</abstract>
<identifier type="citekey">narumanchi-etal-2026-cascade</identifier>
<location>
<url>https://aclanthology.org/2026.findings-aacl.169/</url>
</location>
<part>
<date>2026-11</date>
<extent unit="page">
<start>3376</start>
<end>3391</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T CASCADE: Cognitive Assessment via Sentence Context and Distortion Extraction
%A Narumanchi, Archit
%A Rashidov, Azamat
%A Park, Jong C.
%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 narumanchi-etal-2026-cascade
%X Cognitive distortions are unhelpful thinking patterns central to Cognitive Behavioral Therapy (CBT) and closely linked to mental health outcomes. Despite their context-dependent nature, studies at the intersection of Natural Language Processing and Cognitive Behavioral Therapy are centered largely around sentence-level utterances, ignoring the document-level complexity of real clinical text. To this end, we propose a two-stage rank-then-classify cascade that decomposes document-level multi-label cognitive distortion detection into (1) a sentence-ranking stage that identifies the most likely distorted sentences, and (2) a multi-label classification stage that assigns distortion subtypes to the selected sentences by incorporating their surrounding interleaved context. While the pipeline stages are trained on separate out-of-domain datasets, our model matches state-of-the-art performance on the CBT-CD benchmark of 166 expert-annotated clinical vignettes. Operating with only 368M parameters, our system achieves a 51.8% weighted F1 (9-class), surpassing massive large language model baselines such as Mistral-7B-v0.3 (46.0%) and performing on par with state-of-the-art models such as Gemma 4 26B A4B (50.8%).
%U https://aclanthology.org/2026.findings-aacl.169/
%P 3376-3391
Markdown (Informal)
[CASCADE: Cognitive Assessment via Sentence Context and Distortion Extraction](https://aclanthology.org/2026.findings-aacl.169/) (Narumanchi et al., Findings 2026)
ACL