This paper describes our "breaker" submission to the 2017 EMNLP "Build It Break It" shared task on sentiment analysis. In order to cause the "builder" systems to make incorrect predictions, we edited items in the blind test data according to linguistically interpretable strategies that allow us to assess the ease with which the builder systems learn various components of linguistic structure. On the whole, our submitted pairs break all systems at a high rate (72.6%), indicating that sentiment analysis as an NLP task may still have a lot of ground to cover. Of the breaker strategies that we consider, we find our semantic and pragmatic manipulations to pose the most substantial difficulties for the builder systems.
Some accounts of presupposition projection predict that content's consistency with the Common Ground influences whether it projects (e.g., Heim 1983, Gazdar 1979a,b). I conducted an experiment to test whether Common Ground information about the speaker's social identity influences projection of clausal complement contents (CCs). Participants rated the projection of CCs conveying liberal or conservative political positions when the speaker was either Democrat- or Republican-affiliated. As expected, CCs were more projective when they conveyed political positions consistent with the speaker's political affiliation: liberal CCs were more projective with Democrat compared to Republican speakers, and conservative CCs were more projective with Republican compared to Democrat speakers. In addition, CCs associated with factive predicates (e.g., know) were more projective than those associated with non-factive predicates (e.g., believe). These findings suggest that social meaning influences projective meaning and that social meaning is constrained by semantic meaning, in line with previous research on the relation between other levels of linguistic structure/perception and social information.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.