“Lingmotif2: Universal Platform for Sentiment Analysis” (Ref. FFI2016-78141-P)

This project falls within the scope of the Digital Humanities and is meant to ensure follow-through on the results obtained in our previous Sentiment Analysis project (Lingmotif), which resulted in the creation of a powerful multilingual, multi-platform Sentiment Analysis tool, usable both with general-language and domain-specific texts. The range of applications of this tool goes beyond the usual Natural Language Processing features. Apart from the sentiment score that quantitatively qualifies the input text, Lingmotif provides a detailed, qualitative analysis, as well as a visualization of its sentiment profile. These features makes it a useful tool for research purposes in many fields that involve the study of evaluative language.
Unlike other existing tools, centered on specific kinds of texts (mostly user reviews), the adaptable design of Lingmotif enables users to analyse texts of all kinds and sizes, producing acceptable results across all of them. The real difference lies in the quality of the lexical resources contained in the application, which is the actual sentiment engine. These resources, on the other hand, do not limit the capabilities, since they can be modified in many ways, including the expansion of the predefined context rules (or Context Valence Shifters), as well as the dictionaries themselves. The global sentiment score for the text is computed as a function of the positive and negative scores for the individual items in the text. It is also able to take a set of documents as input, in which case it produces both the above-mentioned analysis, for each document, and a quantitative analysis of the collection. This allows, for example, processing time series, which opens a wide range of possibilities.
In this project we seek to inquire further in the nature of evaluative language and also facilitate this task to others. We will do this by pursuing four general objectives. First, we will revise the current method for computing the overall sentiment score, using more sophisticated statistical techniques. To achieve this, we will use corpus samples from which to derive a number of baselines for different text genres. The sentiment scores calculated on deviations from these baselines will be directly comparable across a diversity of text types and genres. Second, we will expand the input languages that Lingmotif is currently able to analyse (English and Spanish) to two new languages: German and French. Although acquiring the lexical data for a language is far from trivial, we will leverage on the experience acquired during past projects, reusing existing resources, and completing them by means of tried-and-tested corpus-based techniques. Third, we will work to generate the and maintain solid user base by making the application known to the relevant communities and creating the infrastructure to provide support. Finally, we will debug and enhance the application itself, in order to offer a better user experience, with new functionalities, such as the analysis of online feeds (Twitter, RSS) or selectable user profiles (e.g., commercial and academic).
We thus aim to produce a high-coverage, high-value tool for a vast array of users, from researchers in diverse areas, such as discourse analysis, rhetoric, psychology, or marketing, to professionals in the language industries and others.