News

  • November 10, 2016: Student travel grants sponsored by NSF announced
  • November 10, 2016: First call for papers

About the Workshop

During the last decade, semantic representation of text has focused on extracting propositional meaning, i.e., capturing who does what to whom, how, when and where. Several corpora are available, and existing tools extract this kind of knowledge, e.g., role labelers trained on PropBank or NomBank. Nevertheless, most current representations tend to disregard significant meaning encoded in human language. For example, sentences 1-2 below share the same argument structure regarding verb contracted, but do not convey the overall meaning. While in the first example John contracting the disease is factual, in the second it is not:

  1. John likely contracted the disease when a mouse bit him in the Adirondacks.
  2. John never contracted the disease although a mouse bit him in the Adirondacks.

In order to truly capture what these sentences mean, aspects of meaning that go beyond identifying events and their roles (e.g., uncertainty, negation and attribution) must be taken into account.

The Workshop on Computational Semantics Beyond Events and Roles focuses on a broad range of semantic phenomena that lays beyond the identification and linking of eventualities and their semantic arguments with relations such as agent (who), theme (what) and location (where), here so called SemBEaR.

SemBEaR is pervasive in human language and, while studied from a theoretical perspective, computational models are still scarce. Humans use language to describe events that do not correlate with a real situation in the world. They express desires, intentions and plans, and also discuss events that did not happen or are unlikely to happen. Events are often described hypothetically, and speculation can be used to explain why something is a certain way without a strong commitment. Humans do not always (want to) tell the (whole) truth: they may use deception to hide lies. Devices such as irony and sarcasm are employed to play with words so that what is said is not what is meant. Finally, humans not only describe their personal views or experiences, but also attribute statements to others. These phenomena are not exclusive of opinionated texts. They are ubiquitous in language, including scientific works (Hyland 1998) and news as exemplified below:

  • Female leaders might have avoided world wars.
  • Political experts speculate that Donald Trump's meltdown is beginning.
  • Infected people typically don't become contagious until they develop symptoms.
  • Medical personnel can be infected if they don't use protective gear, such as surgical masks and gloves.
  • You can only catch Ebola from coming into direct contact with the bodily fluids of someone who has the disease and is showing symptoms.
  • We have never seen a human virus change the way it is transmitted.
  • The government did not release the files until 1998.

In its 2017 edition, the SemBEaR workshop aims at bringing together scientists working on these type of semantic phenomena within computational semantics. The goal is to attract researchers interested in theoretical frameworks, annotation schemas, modeling and implementing real systems, as well as analyzing the impact of SemBEaR in NLP applications. The workshop also aims at building a bridge between theoretical and computational linguistics.

Topics of interest can be found in the Call for Papers.

SemBEaR-2017 is a follow-up of four previous workshops: ExProM 2016, ExProM 2015, ExProM 2012, and Negation and Speculation in Natural Language Processing (NeSp-NLP 2010).

Organization


Questions? Contact the organizers.