Analysis of spontaneous speech in Parkinson's disease by natural language processing

Unveiling the Diagnostic Potential of Linguistic Markers in Identifying Individuals with Parkinson’s Disease through Artificial Intelligence: A Systematic Review

“…In the selected studies, linguistic measures used to train the ML model were extracted from the transcriptions of spontaneous speech, connected speech, and retelling tasks and analyzed using statistical and word-embedding techniques. More specifically, in most cases, participants were asked to talk about their daily routine [25,[38][39][40][43][44][45]. Other studies used semi-structured interviews where subjects were questioned about their favorite food and their interests, jobs, and families to elicitate spontaneous discourse [38,41].…”

“…More specifically, in most cases, participants were asked to talk about their daily routine [25,[38][39][40][43][44][45]. Other studies used semi-structured interviews where subjects were questioned about their favorite food and their interests, jobs, and families to elicitate spontaneous discourse [38,41]. A study conducted by Jessiman and colleagues [46] used two different tasks: in the first, participants were asked to produce scripts for 10 selected everyday activities, and in the second, participants were asked to provide directions to arrive at 5 familiar destinations.…”

“…NLP combines computational linguistics with statistical, ML, and deep learning models. In the selected studies, several NPL methods were used: (1) tokenization and part-of-speech tagging, consisting of splitting sentences or words into semantically useful units called "tokens" and giving each token a label indicating its speech category [25,38,41,[43][44][45][46]; (2) dependency parsing, used to analyze the grammatical structure in a sentence and define the relationships between words [38,46]; (3) named entity recognition, employed to detect and categorize important information in texts; (4) semantic similarity between sentences, used to exploit word embeddings, that are representations of words in the form of numerical vectors capable of capturing context-dependent linguistic similarities [25,40,42,44]; (5) sentiment analysis used to identify positive or negative emotions in a text [46].…”

“…The most frequently used ML algorithm for automated classification was the Support Vector Machine (SVM), used in 8 out of 10 papers [25,38,[40][41][42][43][44][45]. Other ML algorithms frequently used were Random Forest (RF) [41,44,46], 1D-Convolutional Neural Networks (1D-CNN) with fully connected layer [39,40], Logistic Regression (LR) and Stochastic Gradient Descent (SGD) [43], K-nearest neighbors (KNN) [25,41], Boosting (BG) and Extreme Gradient Boosting (XGBoost) [41], Ada Boost, Linear Discriminant Analysis (LDA), and Naive Bayes (NB) [25].…”

Sustainable Urban Green Blue Space (UGBS) and Public Participation: Integrating Multisensory Landscape Perception from Online Reviews

“…Currently, NLP is widely used in many fields, such as machine translation, spam detection, and medical information extraction [45,86,87]. However, little research has been conducted on public opinions on the UGBS.…”