On the Stability of Some Idiolectal Features

Abstract

Recently due to the development of Internet communication problems of author identification, author profiling (i.e. text-based demographics and personality prediction), deception detection in text, etc. have become burning issues. However, the choice of linguistic parameters for addressing these problems is still an open question. We argue that the level of the stability of these parameters in the idiolect of the individual has to be taken into account while choosing them for the particular task. In this study we are investigating the level of stability of linguistic parameters in an idiolect that are extracted from Russian texts using the Linguistic Inquiry and Word Count (LIWC) software. LIWC is the most commonly used text analysis program in studies related to authorship profiling, deception detection, mental health diagnostics, etc. However, up until recently the stability of LIWC variables in texts by the same individual on different topics etc. has not been addressed. For the first time the parameters of three groups have been identified based on their stability in idiolect computed using a coefficient of variation. The study was performed on the material of unedited informal Russian written texts. The principles of choosing the linguistic parameters for a range of text analysis tasks depending on the level of variation of the parameters in the idiolect have also been set forth.