A Sentiment Analysis-Based Model for Controlling Tone in Written Compositio

The profound material value of the application of natural language processing (NLP) techniques to written composition for various purposes in everyday life has not yet been fully realized. Most of the ideation required in written composition is a function of individual verbal intuition and a command of language formed through varying degrees of personal experience. This creative process could be greatly improved in cases where some input provided by aids to written composition which employ relevant NLP techniques might provide meaningful benefit.;Though many word processing products presently offer some form of spelling or grammar assistance (such as automatic language correction found in the popular Google Docs and Microsoft Word products), no features in such products address a user's interest in affecting the tone present in a written composition. With the recent advent of general-purpose, public-facing Web-based utilities and non-Web-based libraries existing within the subdomain of NLP known as sentiment analysis (SA), the possibilities for integrating novel tone-oriented systems in various contexts for the automation of some aspects of verbal ideation are far-reaching. The domain known as systemic functional linguistics (SFL) is useful in providing a theoretical context for such possibilities through its emphasis on language as a "system of choice".;To that end, this paper will present a model for a system which employs a method of synonymous substitution (SS), given an input document consisting of natural language content. This SS system involves sentiment characterization of the document's tokens (i.e., words which have been split by their positions within a greater linguistic structure), sentences, and the whole document by the sentiment value of each structure on a polar "negative"-to-"positive" scale as follows:;• Very negative (e.g., "hate").;• Negative (e.g., "inconsequential").;• Neutral (e.g., "intention").;• Positive (e.g., "courageous").;• Very positive (e.g., "excellent").;("Excellent" is not the opposite of "hate"; it is an example of "very positive" at the far end of the spectrum from "very negative".) Following this process is a programmatic generation and presentation of token-level synonyms as substitution candidates (with each synonym having also been characterized by its tone).;The ultimate purpose of this model is to provide an application experience in which sentiment-tagged synonyms are presented for the individual tokens of an input document in order to allow a user to effect meaningful tone changes pursuant to rhetorical interests. An implementation of this model will be presented as proof-of-concept.