@inproceedings{li-etal-2025-bigger,
title = "Bigger But Not Better: Small Neural Language Models Outperform {LLM}s in Detection of Thought Disorder",
author = "Li, Changye and
Xu, Weizhe and
Pakhomov, Serguei and
Bradley, Ellen and
Ben-Zeev, Dror and
Cohen, Trevor",
editor = "Zirikly, Ayah and
Yates, Andrew and
Desmet, Bart and
Ireland, Molly and
Bedrick, Steven and
MacAvaney, Sean and
Bar, Kfir and
Ophir, Yaakov",
booktitle = "Proceedings of the 10th Workshop on Computational Linguistics and Clinical Psychology (CLPsych 2025)",
month = may,
year = "2025",
address = "Albuquerque, New Mexico",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2025.clpsych-1.8/",
doi = "10.18653/v1/2025.clpsych-1.8",
pages = "90--105",
ISBN = "979-8-89176-226-8",
abstract = "Disorganized thinking is a key diagnostic indicator of schizophrenia-spectrum disorders. Recently, clinical estimates of the severity of disorganized thinking have been shown to correlate with measures of how difficult speech transcripts would be for large language models (LLMs) to predict. However, LLMs' deployment challenges {--} including privacy concerns, computational and financial costs, and lack of transparency of training data {--} limit their clinical utility. We investigate whether smaller neural language models can serve as effective alternatives for detecting positive formal thought disorder, using the same sliding window based perplexity measurements that proved effective with larger models. Surprisingly, our results show that smaller models are more sensitive to linguistic differences associated with formal thought disorder than their larger counterparts. Detection capability declines beyond a certain model size and context length, challenging the common assumption of ``bigger is better'' for LLM-based applications. Our findings generalize across audio diaries and clinical interview speech samples from individuals with psychotic symptoms, suggesting a promising direction for developing efficient, cost-effective, and privacy-preserving screening tools that can be deployed in both clinical and naturalistic settings."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="li-etal-2025-bigger">
<titleInfo>
<title>Bigger But Not Better: Small Neural Language Models Outperform LLMs in Detection of Thought Disorder</title>
</titleInfo>
<name type="personal">
<namePart type="given">Changye</namePart>
<namePart type="family">Li</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Weizhe</namePart>
<namePart type="family">Xu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Serguei</namePart>
<namePart type="family">Pakhomov</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ellen</namePart>
<namePart type="family">Bradley</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Dror</namePart>
<namePart type="family">Ben-Zeev</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Trevor</namePart>
<namePart type="family">Cohen</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2025-05</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Proceedings of the 10th Workshop on Computational Linguistics and Clinical Psychology (CLPsych 2025)</title>
</titleInfo>
<name type="personal">
<namePart type="given">Ayah</namePart>
<namePart type="family">Zirikly</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Andrew</namePart>
<namePart type="family">Yates</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Bart</namePart>
<namePart type="family">Desmet</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Molly</namePart>
<namePart type="family">Ireland</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Steven</namePart>
<namePart type="family">Bedrick</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Sean</namePart>
<namePart type="family">MacAvaney</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Kfir</namePart>
<namePart type="family">Bar</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yaakov</namePart>
<namePart type="family">Ophir</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Albuquerque, New Mexico</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
<identifier type="isbn">979-8-89176-226-8</identifier>
</relatedItem>
<abstract>Disorganized thinking is a key diagnostic indicator of schizophrenia-spectrum disorders. Recently, clinical estimates of the severity of disorganized thinking have been shown to correlate with measures of how difficult speech transcripts would be for large language models (LLMs) to predict. However, LLMs’ deployment challenges – including privacy concerns, computational and financial costs, and lack of transparency of training data – limit their clinical utility. We investigate whether smaller neural language models can serve as effective alternatives for detecting positive formal thought disorder, using the same sliding window based perplexity measurements that proved effective with larger models. Surprisingly, our results show that smaller models are more sensitive to linguistic differences associated with formal thought disorder than their larger counterparts. Detection capability declines beyond a certain model size and context length, challenging the common assumption of “bigger is better” for LLM-based applications. Our findings generalize across audio diaries and clinical interview speech samples from individuals with psychotic symptoms, suggesting a promising direction for developing efficient, cost-effective, and privacy-preserving screening tools that can be deployed in both clinical and naturalistic settings.</abstract>
<identifier type="citekey">li-etal-2025-bigger</identifier>
<identifier type="doi">10.18653/v1/2025.clpsych-1.8</identifier>
<location>
<url>https://aclanthology.org/2025.clpsych-1.8/</url>
</location>
<part>
<date>2025-05</date>
<extent unit="page">
<start>90</start>
<end>105</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Bigger But Not Better: Small Neural Language Models Outperform LLMs in Detection of Thought Disorder
%A Li, Changye
%A Xu, Weizhe
%A Pakhomov, Serguei
%A Bradley, Ellen
%A Ben-Zeev, Dror
%A Cohen, Trevor
%Y Zirikly, Ayah
%Y Yates, Andrew
%Y Desmet, Bart
%Y Ireland, Molly
%Y Bedrick, Steven
%Y MacAvaney, Sean
%Y Bar, Kfir
%Y Ophir, Yaakov
%S Proceedings of the 10th Workshop on Computational Linguistics and Clinical Psychology (CLPsych 2025)
%D 2025
%8 May
%I Association for Computational Linguistics
%C Albuquerque, New Mexico
%@ 979-8-89176-226-8
%F li-etal-2025-bigger
%X Disorganized thinking is a key diagnostic indicator of schizophrenia-spectrum disorders. Recently, clinical estimates of the severity of disorganized thinking have been shown to correlate with measures of how difficult speech transcripts would be for large language models (LLMs) to predict. However, LLMs’ deployment challenges – including privacy concerns, computational and financial costs, and lack of transparency of training data – limit their clinical utility. We investigate whether smaller neural language models can serve as effective alternatives for detecting positive formal thought disorder, using the same sliding window based perplexity measurements that proved effective with larger models. Surprisingly, our results show that smaller models are more sensitive to linguistic differences associated with formal thought disorder than their larger counterparts. Detection capability declines beyond a certain model size and context length, challenging the common assumption of “bigger is better” for LLM-based applications. Our findings generalize across audio diaries and clinical interview speech samples from individuals with psychotic symptoms, suggesting a promising direction for developing efficient, cost-effective, and privacy-preserving screening tools that can be deployed in both clinical and naturalistic settings.
%R 10.18653/v1/2025.clpsych-1.8
%U https://aclanthology.org/2025.clpsych-1.8/
%U https://doi.org/10.18653/v1/2025.clpsych-1.8
%P 90-105
Markdown (Informal)
[Bigger But Not Better: Small Neural Language Models Outperform LLMs in Detection of Thought Disorder](https://aclanthology.org/2025.clpsych-1.8/) (Li et al., CLPsych 2025)
ACL