@inproceedings{hoogland-etal-2026-multilingual,
title = "Multilingual Cognitive Impairment Detection in the Era of Foundation Models",
author = "Hoogland, Damar and
Koloski, Boshko and
Caporusso, Jaya and
Kolenik, Tine and
Pollak, Senja and
Manouilidou, Christina and
Purver, Matthew",
editor = {Kokkinakis, Dimitrios and
Themistocleous, Charalambos and
Dias, Ga{\"e}l and
Fraser, Kathleen C. and
{\"O}hman, Fredrik and
Pais, Sebasti{\~a}o},
booktitle = "Proceedings of the Sixth Resources and {P}rocess{I}ng of linguistic, para-linguistic and extra-linguistic Data from people with various forms of cognitive/psychiatric/developmental impairments in cooperation with the {MENTAL}.ai consortium",
month = may,
year = "2026",
address = "Palma, Mallorca, Spain",
publisher = "European Language Resources Association (ELRA)",
url = "https://aclanthology.org/2026.rapid-1.1/",
doi = "10.63317/487aco6yfwyv",
pages = "1--12",
abstract = "We evaluate cognitive impairment (CI) classification from transcripts of speech in English, Slovene, and Korean. We compare zero-shot large language models (LLMs) used as direct classifiers under three input settings{---}transcript-only, linguistic-features-only, and combined{---}with supervised tabular approaches trained under a leave-one-out protocol. The tabular models operate on engineered linguistic features, transcript embeddings, and early or late fusion of both modalities. Across languages, zero-shot LLMs provide competitive no-training baselines, but supervised tabular models generally perform better, particularly when engineered linguistic features are included and combined with embeddings. Few-shot experiments focusing on embeddings indicate that the value of limited supervision is language-dependent, with some languages benefiting substantially from additional labelled examples while others remain constrained without richer feature representations. Overall, the results suggest that, in small-data CI detection, structured linguistic signals and simple fusion-based classifiers remain strong and reliable signals."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="hoogland-etal-2026-multilingual">
<titleInfo>
<title>Multilingual Cognitive Impairment Detection in the Era of Foundation Models</title>
</titleInfo>
<name type="personal">
<namePart type="given">Damar</namePart>
<namePart type="family">Hoogland</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Boshko</namePart>
<namePart type="family">Koloski</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jaya</namePart>
<namePart type="family">Caporusso</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Tine</namePart>
<namePart type="family">Kolenik</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Senja</namePart>
<namePart type="family">Pollak</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Christina</namePart>
<namePart type="family">Manouilidou</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Matthew</namePart>
<namePart type="family">Purver</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2026-05</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Proceedings of the Sixth Resources and ProcessIng of linguistic, para-linguistic and extra-linguistic Data from people with various forms of cognitive/psychiatric/developmental impairments in cooperation with the MENTAL.ai consortium</title>
</titleInfo>
<name type="personal">
<namePart type="given">Dimitrios</namePart>
<namePart type="family">Kokkinakis</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Charalambos</namePart>
<namePart type="family">Themistocleous</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Gaël</namePart>
<namePart type="family">Dias</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Kathleen</namePart>
<namePart type="given">C</namePart>
<namePart type="family">Fraser</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Fredrik</namePart>
<namePart type="family">Öhman</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Sebastião</namePart>
<namePart type="family">Pais</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>European Language Resources Association (ELRA)</publisher>
<place>
<placeTerm type="text">Palma, Mallorca, Spain</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
</relatedItem>
<abstract>We evaluate cognitive impairment (CI) classification from transcripts of speech in English, Slovene, and Korean. We compare zero-shot large language models (LLMs) used as direct classifiers under three input settings—transcript-only, linguistic-features-only, and combined—with supervised tabular approaches trained under a leave-one-out protocol. The tabular models operate on engineered linguistic features, transcript embeddings, and early or late fusion of both modalities. Across languages, zero-shot LLMs provide competitive no-training baselines, but supervised tabular models generally perform better, particularly when engineered linguistic features are included and combined with embeddings. Few-shot experiments focusing on embeddings indicate that the value of limited supervision is language-dependent, with some languages benefiting substantially from additional labelled examples while others remain constrained without richer feature representations. Overall, the results suggest that, in small-data CI detection, structured linguistic signals and simple fusion-based classifiers remain strong and reliable signals.</abstract>
<identifier type="citekey">hoogland-etal-2026-multilingual</identifier>
<identifier type="doi">10.63317/487aco6yfwyv</identifier>
<location>
<url>https://aclanthology.org/2026.rapid-1.1/</url>
</location>
<part>
<date>2026-05</date>
<extent unit="page">
<start>1</start>
<end>12</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Multilingual Cognitive Impairment Detection in the Era of Foundation Models
%A Hoogland, Damar
%A Koloski, Boshko
%A Caporusso, Jaya
%A Kolenik, Tine
%A Pollak, Senja
%A Manouilidou, Christina
%A Purver, Matthew
%Y Kokkinakis, Dimitrios
%Y Themistocleous, Charalambos
%Y Dias, Gaël
%Y Fraser, Kathleen C.
%Y Öhman, Fredrik
%Y Pais, Sebastião
%S Proceedings of the Sixth Resources and ProcessIng of linguistic, para-linguistic and extra-linguistic Data from people with various forms of cognitive/psychiatric/developmental impairments in cooperation with the MENTAL.ai consortium
%D 2026
%8 May
%I European Language Resources Association (ELRA)
%C Palma, Mallorca, Spain
%F hoogland-etal-2026-multilingual
%X We evaluate cognitive impairment (CI) classification from transcripts of speech in English, Slovene, and Korean. We compare zero-shot large language models (LLMs) used as direct classifiers under three input settings—transcript-only, linguistic-features-only, and combined—with supervised tabular approaches trained under a leave-one-out protocol. The tabular models operate on engineered linguistic features, transcript embeddings, and early or late fusion of both modalities. Across languages, zero-shot LLMs provide competitive no-training baselines, but supervised tabular models generally perform better, particularly when engineered linguistic features are included and combined with embeddings. Few-shot experiments focusing on embeddings indicate that the value of limited supervision is language-dependent, with some languages benefiting substantially from additional labelled examples while others remain constrained without richer feature representations. Overall, the results suggest that, in small-data CI detection, structured linguistic signals and simple fusion-based classifiers remain strong and reliable signals.
%R 10.63317/487aco6yfwyv
%U https://aclanthology.org/2026.rapid-1.1/
%U https://doi.org/10.63317/487aco6yfwyv
%P 1-12
Markdown (Informal)
[Multilingual Cognitive Impairment Detection in the Era of Foundation Models](https://aclanthology.org/2026.rapid-1.1/) (Hoogland et al., RaPID 2026)
ACL
- Damar Hoogland, Boshko Koloski, Jaya Caporusso, Tine Kolenik, Senja Pollak, Christina Manouilidou, and Matthew Purver. 2026. Multilingual Cognitive Impairment Detection in the Era of Foundation Models. In Proceedings of the Sixth Resources and ProcessIng of linguistic, para-linguistic and extra-linguistic Data from people with various forms of cognitive/psychiatric/developmental impairments in cooperation with the MENTAL.ai consortium, pages 1–12, Palma, Mallorca, Spain. European Language Resources Association (ELRA).