The growing global burden of dementia underscores the urgent need for scalable, objective screening tools. While traditional diagnostic methods rely on subjective assessments, advances in natural language processing offer promising alternatives. In this study, we compare two classes of language models—encoder-based pretrained language models (PLMs) and autoregressive large language models (LLMs) for detecting cognitive impairment from narrative speech. Using the DementiaBank Pitt Corpus and the widely used Cookie Theft picture description task, we evaluate BERT as a representative PLM alongside GPT-2, GPT-3.5 Turbo, GPT-4, and LLaMA-2 as LLMs. Although all models are pretrained, we distinguish PLMs and LLMs based on their architectural differences and training paradigms. Our findings reveal that BERT outperforms all other models, achieving 86% sensitivity and 95% specificity. LLaMA-2 follows closely, while GPT-4 and GPT-3.5 underperform in this structured classification task. Interestingly, LLMs demonstrate complementary strengths in capturing narrative richness and subtler linguistic features. These results suggest that hybrid modeling approaches may offer enhanced performance and interpretability. Our study highlights the potential of language models as digital biomarkers and lays the groundwork for scalable, AIpowered tools to support early dementia screening in clinical practice.

The choice of attributes in the multi-attribute decision-making process becomes frequently uncertain because of the diverse degree of preference for alternatives. These are assessed utilizing human decisions and linguistic terms that can be utilized for a more adaptable and delicate assessment. The present article illustrates a multi-attribute decision-making (MADM) process, named the exponential technique for order of preference by similarity to an ideal solution (Exp-TOPSIS), considering the selection of attributes with existing uncertainty. Another three notable multi-attribute decision-making (MADM) processes, termed as multi-attribute utility theory (MAUT), elimination and choice expressing reality method (ELECTRE), and the technique for order of preference by similarity to an ideal solution (TOPSIS) are utilized to present a comparison with the proposed methodology by proposing a mathematical model for a solid transportation problem intending to minimize carbon emissions under an uncertain environment. The uncertainty theory, which depends on human conviction degree, is utilized to define the uncertain parameters of the model related to the problem. Applying the proposed one and the other three multi-attribute decision-making processes, the best emission factors are observed to mitigate the carbon emissions from the transport sectors. In this context, the proposed method has some advantages over the existing techniques in selecting the emission factors. All four MADM approaches with different weights have been tested to choose the best five attributes among nine options to be utilized in the mathematical model to minimize the total carbon emission ejection from transportation. In every case, the obtained result states that the proposed Exp-TOPSIS gives the minimum carbon emissions in a range of 2100–2500 units. LINGO 13.0 solver is used to address the deterministic solid transportation problem, and finally, this study presents some investigations on the selection of carbon emission factors and future utilization of the proposed multi-attribute decision-making process.