Search for: All records

This paper surveys 451 Americans about how they view and interpret three semantically indeterminate progressive political slogans: #BelieveWomen, #DefundThePolice, and #FreePalestine. In each case, most people who agree with the slogan interpret it to express a moderate position, while most people who disagree take it to describe a more extreme position – which is indeed endorsed by a minority of those who agree with the slogan. These results show that online political discourse can foment both false controversy and false consensus. Because liberals tend to interpret these slogans moderately, while conservatives are more likely to interpret them as extreme, these results further suggest that people may choose their interpretation of a slogan to foreground the issues that they see as problems, and/or to justify their preexisting attitude towards the movement it champions. This paper brings together linguistics and political science to illuminate miscommunication in public discourse 

Which adjectives tend to occur as attributive ("the cute/red dress") versus predicative ("the dress is cute/red") and why? Building on findings from Wiegand et al. (2013. ) and Vartiainen (2013), this paper argues that subjective adjectives such as "cute" tend to be placed in predicative position not just because they often describe discourse-new information, but because this position serves to foreground information that the hearer may disagree with. This claim is supported using data from the Corpus of Contemporary American English (Davies, Mark. 2008) combined with human annotations for subjectivity from Scontras et al. (2017) et seq.; and data from image captions versus descriptions (for seeing versus low-vision people) from the National Gallery of Art. A production experiment manipulates the discourse context to further show that adjectives tend to be placed in predicative position when they express controversial information. Overall, this paper explores how the lexical semantics of adjectives shapes the pragmatic contexts in which they tend to be used, which in turn shapes the syntax of the sentences using them. 

This paper sets out to explain why the verb CAUSE tends to occur with negative-sentiment complements (CAUSE DAMAGE, CAUSE PROBLEMS), as observed by Stubbs 1995. Formalized using causal models (Pearl 2000, Halpern & Pearl 2005, Schulz 2011), the analysis hinges on the asymmetric inference patterns licensed by necessary versus sufficient causes in the common scenario where some variables in a causal model remain uncertain. States of certainty/uncertainty are captured by subdividing the traditional definitions of necessity and sufficiency into a local version (all other variables fixed at particular values) and a global version (all other variables unsettled). C CAUSES E is argued to entail that that C is locally sufficient for E, and to implicate that C is at least possibly locally necessary for E. With this definition, it is shown that C CAUSES E can be truthfully applied to more uncertain contexts when C is a globally sufficient cause of E rather than a globally necessary one. CAUSE thus tends to occur with outcomes depending on a single globally sufficient cause -- outcomes which are moreover shown to be negative in sentiment, reflecting the independently motivated “Anna Karenina Principle” that bad outcomes tend to require single sufficient causes, thus indirectly explaining why CAUSE prefers negative-sentiment complements. The meaning and collocational sentiment of CAUSE are used to illuminate one another. 

Abstract We present an event structure classification empirically derived from inferential properties annotated on sentence- and document-level Universal Decompositional Semantics (UDS) graphs. We induce this classification jointly with semantic role, entity, and event-event relation classifications using a document-level generative model structured by these graphs. To support this induction, we augment existing annotations found in the UDS1.0 dataset, which covers the entirety of the English Web Treebank, with an array of inferential properties capturing fine-grained aspects of the temporal and aspectual structure of events. The resulting dataset (available at decomp.io) is the largest annotation of event structure and (partial) event coreference to date.