Evaluative language beyond bags of words: Linguistic insights and computational applications

Abstract

The expression of evaluation and opinion is a central aspect of language. It allows us to convey feelings, assessments of people, situations and objects, and to engage with other opinion holders. The study of evaluation, affect and subjectivity is a multidisciplinary enterprise, including sociology, psychology, economics, linguistics, and computer science. A number of excellent computational linguistics and linguistic surveys of the field exist. As far as we know, no survey has attempted to bring the two disciplines together to show how methods from linguistics can benefit computational sentiment analysis systems. In the survey, we show how incorporating linguistic insights, discourse information and other contextual phenomena, in combination with the statistical exploitation of data can result in an improvement over approaches which take advantage of only one of those perspectives. We first provide a comprehensive introduction to evaluative language from both a linguistic and computational perspective. We argue that the standard computational definition of the concept of evaluation neglects the dynamic nature of evaluation in which the interpretation of a given evaluation depends on linguistic and extra-linguistic contextual factors. We thus propose a new dynamic definition that incorporates update functions which allow for different contextual aspects to be incorporated into the calculation of sentiment for evaluative words or expressions, and can be applied to all discourse levels.We explore each level and highlight which linguistic aspects contribute to accurate extraction of sentiment. This survey is an attempt to show that future breakthroughs in Natural Language Processing applications will require developing synergies between machine learning techniques and bag-of-words representations on the one hand, and in-depth theoretical accounts and detailed analyses of linguistic phenomena on the other hand.