Innovation and challenges of spoken and aphasic speech data: Introducing the CACLA corpora

Aphasia assessment in Modern Greek is hampered by the scarcity of authentic spoken data, annotated resources, and adaptive clinical instruments; SUBTLEX-GR offers frequency counts drawn from translated subtitles, yet clinicians still lack the psycholinguistic norms routinely available for English and other major languages. Computational linguistics provides a bridge between linguistic theory and clinical practice by allowing theoretical assumptions to be tested through computational implementation, and we exploit this bridge to address the gap. We present the CACLA (Clinical Application of Computational Linguistics for Aphasia) corpora, comprising: (a) GRTVScripts (Greek television scripts), (b) GRNewsTexts (broadcast news files), (c) ENGTVGreekSubtitles (Greek subtitles), and (d) AphasiaSpeech (clinical naming and narration data from speakers with aphasia). Using established corpus tools (AntConc, Voyant, Sketch Engine, GPT-4), we derive word frequencies, contextual diversity, Part-of-Speech-specific frequencies, and bigram measures, yielding the comprehensive Modern Greek norms that clinical assessment has so far lacked. These resources directly underpin the development of a computer-adaptive anomia test for Greek, addressing the limitations of current instruments and grounding item selection in both cognitive and psychometric theory. The result is a tangible step towards theoretically grounded, data-driven aphasia evaluation tailored to Greek-speaking patients.