Pain is a private and subjective experience, dependent on a complex cognitive process. These properties complicate the clinician’s essential task of perceiving the patient’s actual experience, given that it is only accessible from the outside through the lens of expressions and behaviors of pain. By modeling this experience through verbal reporting, it may be possible to better objectify the pain perceived by the patient, and thus provide a more adequate, personal treatment. This is motivated by the study of the cognitive process that embeds the painful experience, which determines, namely, that the emotional, psychosocial, and sociocultural dimensions of the subject in pain play a specific part in modulating the perception of pain and corresponding suffering and expression, and the study of the language of pain, which is shown to carry part of this information.
The primary objective of UNBABEL is to evaluate the verbal description of the chronic pain experience of a restricted group of patients in order to develop a computational method to support clinical pain assessment. To this end, three main phases can be outlined, (1) data collection, (2) linguistic analysis, and (3) paralinguistic analysis.
The data collection phase aims at continuously collecting interviews with such patients, attending to the context in which the pain experience is being described, as it is shown to affect both the vocabulary and the way it is reported. Therefore, with respect to the primary objective, the data collection, consisting of an interview and a complementary form per patient, will take place in the medical office immediately after the medical appointment. The set of questions composing the interview is the result of a design process that aimed at obtaining a natural description of the patient’s pain experience, in their own words, but, at the same time, directing it towards the cognitive topics that have been identified as the most relevant for pain assessment. All data will be anonymized and the presentation of research results will never resort to individual references.
In the linguistic analysis phase, syntactic and semantic structures of textual descriptions of pain may yield correlations between the content of these descriptions and other relevant medical or non-medical aspects of the experience. This includes the identification of the most significant descriptors or qualifying attributes, the aggregation of descriptions focusing on the same or similar concepts, sentiment analysis, and regression of any value from a description. This analysis may be performed with a multitude of methods and models.
The paralinguistic analysis phase, methods to evaluate the patient’s verbalization of pain without explicitly considering the actual semantic or structural contents of the description. Thus, the speech signal can be processed and relevant information not present in the textual transcription can be extracted, which will allow to further characterize the patient and infer possible states.