Postdoctoral Research Fellow
Language Technology Group
Department of Informatics
University of Oslo

phone: +47 2284 0849
email: jread@ifi.uio.no

  Flickr
  Google Scholar
  Mendeley

Research

I am a postdoctoral research fellow in the Department of Informatics at the University of Oslo. I am part of the Logic and Natural Language research group.

My research interests focus on the use of machine-learning techniques for natural language processing (NLP) tasks including: analysis of sentiment, opinions and emotions, resolution of speculation/negation scope, and the development of language resources. Presently I carry out research within the context of the WeSearch project, which explores the development and application of parsing technologies to natural language processing applications within the context of user-generated content, covering text types as diverse as academic articles and Web forums, with a view to applications such as semantic search and sentiment analysis.

Most recently I have investigated the application of support vector machines to analyse speculation and negation in biomedical academic literature. This approach uses constituents in trees generated licensed by the English Resource Grammar to generate features for a linear ranking function to predict the scope of speculation or negation. When combined with a system for cue detection and the scope predictions of a rule-based approach the ranker achieves the best results yet published (on both speculation annotations from the evaluation data of the CoNLL-2010 Shared Task, and on negation annotations from the BioScope corpus).

My doctoral dissertation, advised by Professor John Carroll, explored some drawbacks of using supervised techniques for sentiment analysis, showing that strong performance is dependent on a good match between the training and testing data with respect to domain, topic and time-period. This makes supervised learning impractical for sentiment analysis for applications involving a broad range of text types. Consequently, I investigated weakly supervised methods for sentiment analysis, wherein the sentiment of a text is estimated according to its words' similarity with prototypical examples of positivity and negativity. The second part of my dissertation extended the breadth of research in sentiment analysis beyond the discrimination of positivity and negativity, by considering the different types of textual opinions and how they are expressed, as described by the functional linguistic theory of Appraisal. As part of this research I constructed a corpus of book reviews in which instances of Appraisal were independently annotated by two coders. This corpus provided the first quantitative evaluation of the Appraisal theory, which indicated that while aspects of the theory are difficult, many expressions are readily identifiable. I evaluated the weakly-supervised techniques in classifying and extracting such expressions, finding reasonable precision and recall given the complexity of the task.