SEQUENCE PACKAGE ANALYSIS: A New Natural Language Understanding Method for Intelligent Mining of Recordings of Doctor-Patient Interviews and Health-Related Blogs

Medical histories provide a rich resource for diagnoses and treatment. Similarly, consumers' blog postings on health-related topics offer unique data for medical researchers, practitioners, and pharmacologists. Nevertheless, speech and text analytic programs for mining recordings of patients' histories or consumers' blogs are compromised by the ambiguities, repetitions and ellipses in natural speech, which can be more pronounced when the speaker or blogger is discussing a medical problem. Conventional systems or programs limited to a set of key words and phrases cannot process speech as it actually occurs; if a speaker or blogger fails to use the word(s) found in the speech application's vocabulary, a poor statistical word match (or no match) is given. This paper shows how sequence package analysis is informed by algorithms that can work with, rather than be hindered by, less than perfect natural speech for intelligent mining of doctor-patient recordings and blogs

1. Introduction

Sequence Package Analysis offers a new and advanced Natural Language Understanding (NLU) method whose parsing structures consisting of lexical units and prosodic features are designed to reflect in addition to syntax some of the more complex semantic aspects of communication. A discussion of the methodological basis of Sequence Package Analysis (SPA), formulated by this author, first appeared in the International Journal of Speech Technology [1]. SPA has been referred to by AI researchers as among the more advanced parsing methods for “captioning the text to which data mining is applied” so as to better detect the subtleties of human emotions [2]. By relying on the entire sequence package - a series of related speaking turns and parts of turns - rather than on individual words or combination of words and their attendant stress patterns, SPA can enable a speech system to detect, among other things, the wide range of speakers' emotions in recordings of doctor-patient interviews and health-related blogs.