Subsentential Sentiment on a Shoestring: A Crosslingual Analysis of Compositional Classification

Human Language Technologies: The 2015 Annual Conference of the North American Chapter of the ACL, pages 694–704, Denver, Colorado, May 31 – June 5, 2015. c⃝2015 Association for Computational Linguistics Subsentential Sentiment on a Shoestring: A Crosslingual Analysis of Compositional Classification Michael Haas and Yannick Versley Institute for Computational Linguistics University of Heidelberg {haas,versley}@cl.uni-heidelberg.de Abstract Sentiment analysis has undergone a shift from document-level analysis, where labels ex- presses the sentiment of a whole document or whole sentence, to subsentential approaches, which assess the contribution of individual phrases, in particular including the composi- tion of sentiment terms and phrases such as negators and intensifiers. Starting from a small sentiment treebank mod- eled after the Stanford Sentiment Treebank of Socher et al. (2013), we investigate suitable methods to perform compositional sentiment classification for German in a data-scarce set- ting, harnessing cross-lingual methods as well as existing general-domain lexical resources. 1 Introduction In sentiment classification, we find a general ten- dency from document-level classification towards more fine-grained approaches that yield a more de- tailed appraisal of the judgement performed in the text - in particular, using composition over syntac- tic structure to get a more detailed approach over phrases. For English movie reviews, work using the Stan- ford Sentiment Treebank (SSTb) has shown that such subsentential sentiment information can yield approaches with both very high accuracy (Socher et al., 2013; Dong et al., 2014; Hall et al., 2014) and precise information about the role of each phrase – information which can subsequently used for ex- tracting or summarizing the sentiment expressed in the text. The effort for creating a sentiment treebank such as the SSTb, however, seems prohibitive if we wanted to create such a resource for each pair of relevant domain and language: Compared to document-level annotations for sentiment, which are easy to come by (e.g., star ratings), annotating indi- vidual syntactic phrases requires considerable effort. The main focus of this paper is the question if and how it is possible to reach sensible per- formance for compositional sentiment classifica- tion when we only have limited resources to spend on an in-language, in-domain sentiment treebank. For this goal, we use a new resource, the Hei- delberg Sentiment Treebank (HeiST), which is a German-language counterpart to the Stanford Senti- ment Treebank in the sense that it makes explicit the composition of sentiment expression over syntactic phrases. Our experiments on HeiST provide a di- rect comparison of different techniques for harness- ing cross-lingual, cross-domain, or cross-task infor- mation, and are the first of this kind to specifically target compositional sentiment analysis. Figure 1 (next page) shows a schematic overview of the experiments: beyond supervised baseline ex- periments using SVM classification and a super- vised RNTN model (section 3), we evaluated cross- lingual projection (section 4), lexicon-based ap- proaches (section 5), as well as semi-supervised ap- proaches based on word clusters (section 6). 2 Related Work The starting point for our research is the idea that the sentiment of larger stretches of text can be …

Content truncated to 500 words