People with schizophrenia generally have different speech disorders, such as tangentiality, dissociation of thought, or persevera-tion. These disorders can be classified using the Thought and Language Disorder (TALD) scale, which measures the intensity of the different disorders based on the presence of particular characteristics of the speech. At present, the current practice to measure speech disorders through the TALD scale involves specialized clinicians who administer specific interviews to assess the degree of TALD’s sub-scales with no technological support. Conducting the interviews is a very time consuming activity and an automatic support for the interview administration and the identification of the language disorders may be useful. In this paper, we propose an approach named RoboTald in which a robot is adopted for formulating questions to patients by following the TALD guidelines and to record their answers. The audio files are then automatically transcripted and inputted to a deep-learning-based system aiming at assessing the speech disorders basing on the TALD classification. Faced with the need to improve and speed up the classification and quantification of disorders we adopt NLP models for analyzing the speech characteristics. In particular, the robot dialogues with the patient and collects her speech. Starting from the transcription of patient’s speech, TALD Item measurements are computed by Transformer models, able to recognize the semantic meaning of sentences and to understand the similarity between two inputs using the “Cosine Similarity” technique. This technique identifies the semantic distance between two inputs based on the features extracted from the text provided to the Transformer. The distances are then used to evaluate the elements of the scale by analyzing how far the various sentences deviate from the initial question or how far apart they are from each other. In this preliminary work we focus our attention on two elements of the TALD scale: Tangentiality and Derailment.
Supporting Schizophrenia PatiEnts’ Care wiTh Robotics and Artificial Intelligence
Amaro I.;Francese R.
;Tortora G.;Tucci C.;D'errico L.;
2023
Abstract
People with schizophrenia generally have different speech disorders, such as tangentiality, dissociation of thought, or persevera-tion. These disorders can be classified using the Thought and Language Disorder (TALD) scale, which measures the intensity of the different disorders based on the presence of particular characteristics of the speech. At present, the current practice to measure speech disorders through the TALD scale involves specialized clinicians who administer specific interviews to assess the degree of TALD’s sub-scales with no technological support. Conducting the interviews is a very time consuming activity and an automatic support for the interview administration and the identification of the language disorders may be useful. In this paper, we propose an approach named RoboTald in which a robot is adopted for formulating questions to patients by following the TALD guidelines and to record their answers. The audio files are then automatically transcripted and inputted to a deep-learning-based system aiming at assessing the speech disorders basing on the TALD classification. Faced with the need to improve and speed up the classification and quantification of disorders we adopt NLP models for analyzing the speech characteristics. In particular, the robot dialogues with the patient and collects her speech. Starting from the transcription of patient’s speech, TALD Item measurements are computed by Transformer models, able to recognize the semantic meaning of sentences and to understand the similarity between two inputs using the “Cosine Similarity” technique. This technique identifies the semantic distance between two inputs based on the features extracted from the text provided to the Transformer. The distances are then used to evaluate the elements of the scale by analyzing how far the various sentences deviate from the initial question or how far apart they are from each other. In this preliminary work we focus our attention on two elements of the TALD scale: Tangentiality and Derailment.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
