Background: Lexical access problems of inflected verbs are common in aphasia. Previous research addressed these problems either in purely linguistic terms (e.g., verb movement) or in terms of lexical characteristics (e.g., frequency). We propose a new measure of verb complexity, which combines linguistic and lexical characteristics and is formulated in terms of Shannon’s information theory. Aims: We aim to explore the complexity of individual verbs and verb paradigms and its effect on lexical access, both in unimpaired people and people with aphasia (PWA). We apply information theory to investigate the impact of verb complexity on reaction time (RT) for lexical decision. Methods & Procedures: 20 non-fluent aphasic subjects and 11 age-matched and education-matched peers performed an auditory lexical decision task containing 286 real and 286 phonotactically legal non-word past tense forms (regulars and irregulars). RTs and error rates were measured. Two information-theoretic measures were calculated: inflectional entropy (reflecting probabilistic variability of forms within a given verbal family) and information load (I) (reflecting complexity of an individual verb form). The effect for these and other more traditional measures on RT were measured. Outcomes & Results: Linear mixed model analyses to the data for each group with participant and verb as crossed random effects were performed. Results show that for all groups inflectional entropy had a facilitatory effect on RT. There was a group effect for inflectional entropy indicating that for the patients with aphasia the effect of inflectional entropy was less pronounced. At the same time, I did correlate with latencies for healthy adults but not for individuals with aphasia. Conclusions: Our results demonstrate that the decrease in lexical processing capacity characteristic for PWA has a measurable effect that can be calculated using information theoretical means. According to our model, these individuals have particular difficulties with processing lexical items of higher complexity, as measured by individual I, and benefit less from the support normally provided (in comprehension) by other members of the corresponding lexical network. Finally, the proposed information-theoretic complexity measures, which encompass both frequency effects and linguistic parameters, provide a superior measure of lexical access, and have a better explanatory power for the analyses of access problems found in non-fluent aphasia, compared to analyses based on frequency only.
LINK
Multilevel models using logistic regression (MLogRM) and random forest models (RFM) are increasingly deployed in industry for the purpose of binary classification. The European Commission’s proposed Artificial Intelligence Act (AIA) necessitates, under certain conditions, that application of such models is fair, transparent, and ethical, which consequently implies technical assessment of these models. This paper proposes and demonstrates an audit framework for technical assessment of RFMs and MLogRMs by focussing on model-, discrimination-, and transparency & explainability-related aspects. To measure these aspects 20 KPIs are proposed, which are paired to a traffic light risk assessment method. An open-source dataset is used to train a RFM and a MLogRM model and these KPIs are computed and compared with the traffic lights. The performance of popular explainability methods such as kernel- and tree-SHAP are assessed. The framework is expected to assist regulatory bodies in performing conformity assessments of binary classifiers and also benefits providers and users deploying such AI-systems to comply with the AIA.
DOCUMENT
