Explanation in Computational Stylometry

Abstract

Abstract. Computational stylometry, as in authorship attribution or profiling, has a large potential for applications in diverse areas: literary science, forensics, language psychology, sociolinguistics, even medical diagnosis. Yet, many of the basic research questions of this field are not studied systematically or even at all. In this paper we will go into these problems, and suggest that a reinterpretation of current and historical methods in the framework and methodology of machine learning of natural language processing would be helpful. We also argue for more attention in research for explanation in computational stylometry as opposed to purely quantitative evaluation measures and propose a strategy for data collection and analysis for achieving progress in computational stylometry. We also introduce a fairly new application of computational stylometry in internet security. Meta-knowledge Extraction from TextThe form of a text is determined by many factors. Content plays a role (the topic of a text determines in part its vocabulary), text type (genre, register) is important and will determine part of the writing style, but also psychological and sociological aspects of the author of the text will be sources of stylistic language variation. These psychological factors include personality, mental health, and being a native speaker or not; sociological factors include age, gender, education level, and region of language acquisition.Writing style is a combination of consistent decisions in language production at different linguistic levels (lexical choice, syntactic structures, discourse coherence, ...) that is linked to specific authors or author groups such as male authors or teenage authors. It remains to be seen whether this link is consistent over time and whether there are style features that are unconscious and cannot be controlled, as some researchers have argued. The basic research question for computational stylometry seems then to describe and explain the causal relations between psychological and sociological properties of authors on the one hand, and their writing style on the other. These theories can be used to develop systems that generate text in a particular style, or perhaps more usefully, systems that detect the identity of authors (authorship attribution and verification) or some of their psychological or sociological properties (profiling) from text.A limit hypothesis arising from this definition is that style is unique for an individual, like her fingerprint, earprint or genome. This has been called the human stylome hypothesis:

Function Words in Authorship Attribution. From Black Magic to Theory?

“…The more observations we have available per text, the more trustworthily one can represent it. Character n-grams push this idea even further, simply because texts by definition have more data points for character n-grams than for entire words (Stamatatos, 2009;Daelemans, 2013). Thus the mere number of observations, relatively larger for character n-grams than for function words, might account for their superiority from a purely quantitative perspective.…”

Developing a Cognitive Assistant for the Audit Plan Brainstorming Session

“…Intelligent personal assistants such as Apple's Siri and Amazon's Alexa leverage machine learning techniques, including Deep Neural Networks (DNN), convolutional neural networks (Tang and Lin, 2017), long short-term memory units (Chen et al, 2015), gated recurrent units (Ravuri and Stolcke, 2016), and ngrams (Daelemans, 2013;Levy, 2016) to build smart voice recognition system. (Rybach et al, 2011).…”

The Impact of Oath Writing Style on Stylometric Features and Machine Learning Classifiers

“…Feature selection is most important rather than the choice of machine learning method (Daelemans, 2013). However, classifiers selection is highly related to feature extraction and selection.…”