BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//pretalx//conference-hub.linguistic-society.com//athens-2026//spea
 ker//HGSKMW
BEGIN:VEVENT
UID:pretalx-athens-2026-PABPJS@conference-hub.linguistic-society.com
DTSTART:20260422T083500Z
DTEND:20260422T085000Z
DESCRIPTION:Objectives\nLanguage is dynamic\, meanings converge\, diverge\,
  and form evolving semantic fields. In clinical\nneuropsychology\, however
 \, this variability is typically reduced to fixed categories. Linguistic a
 bility and\nimpairment are commonly assessed using standard neuropsychiatr
 ic instruments such as the Semantic Verbal\nFluency test (SVF)\, the Phono
 logical Verbal Fluency test (FAS)\, and the Boston Naming Test (BNT). Most
 \noften\, responses on these measures are scored dichotomously as correct 
 or incorrect. This binary scoring obscures\nsemantically related\, approxi
 mate\, or deviant responses. The objective of this study is to develop and
  evaluate\na reproducible computational method for continuous\, semantical
 ly informed scoring of these tests in Swedish.\nThe primary research quest
 ion is whether modern vector-based language models can generate stable and
 \ninterpretable continuous semantic scores that capture graded variation b
 eyond binary classification.\n\nMethodology\nThe study applies a computati
 onal linguistic framework grounded in distributional semantics\, where wor
 d\nmeaning is represented as position in a high-dimensional semantic space
 . Anonymised\, synthetically generated\nlexical responses were used to ena
 ble controlled methodological development without sensitive data. Text\npr
 eprocessing\, including normalisation and lemmatisation\, was performed us
 ing tools from Språkbanken’s text\ninfrastructure. Responses and target
  words were represented using Swedish-adapted BERT- based vector\nembeddin
 gs. BERT (“Bidirectional Encoder Representations from Transformers”) i
 s a transformer-based\nlanguage model that learns contextual word represen
 tations by analysing large corpora of text and modelling\nhow words relate
  to surrounding words in both left and right contexts. In this framework\,
  lexical meaning is\nencoded as numerical vectors in a high-dimensional se
 mantic space\, where semantically similar words are\npositioned closer to 
 one another. This representation enables graded measurement of semantic pr
 oximity\nrather than categorical judgments of correctness. For the verbal 
 fluency tests (SVF and FAS)\, semantic\ndispersion was also computed to qu
 antify how responses are distributed within the semantic space. In this\nc
 ontext\, semantic dispersion denotes the quantitative distribution of resp
 onse vectors within a highdimensional embedding space\, operationalised as
  the extent to which lexical items diverge from one another in semantic re
 presentation.\n\nResults\nVector-based representations generated stable an
 d interpretable continuous scores. The method captured\nfine-grained varia
 tion among semantically related responses that is lost under binary scorin
 g. Systematic\ndifferences in response structure were observed across the 
 Boston Naming Test (BNT)\, Semantic Verbal\nFluency (SVF)\, and Phonologic
 al Verbal Fluency (FAS). Linguistic performance could thus be modelled as\
 nmovement within a semantic space rather than as a series of discrete outc
 omes.\n\nDiscussion\nThe study demonstrates the feasibility of continuous 
 semantic scoring for Swedish language assessment. The\nproposed method pro
 vides a methodological foundation for future clinical validation and contr
 ibutes to\nresearch on how meaning is structured and dynamically organised
  in cognitive processes. By reconceptualising\ntest performance as graded 
 semantic movement\, the study advances computational approaches to linguis
 tic\nassessment in neuropsychology. Importantly\, this framework enables t
 he quantification of latent semantic\nstructure in a manner that is theore
 tically grounded\, statistically scalable\, and reproducible across datase
 ts.\nSuch an approach may facilitate more sensitive detection of subtle li
 nguistic deviations\, potentially improve\nearly identification of cogniti
 ve decline and supporting longitudinal monitoring of semantic change over 
 time.
DTSTAMP:20260419T080900Z
LOCATION:Main Auditorium
SUMMARY:Embedding-Based Graded Scoring of Neuropsychological Language Tests
  - Dimitrios Kokkinakis
URL:https://conference-hub.linguistic-society.com/athens-2026/talk/PABPJS/
END:VEVENT
END:VCALENDAR