Ambiguity: Perspectives on Representation and Resolution

 

Language and Computation Courses

Workshop

Ambiguity: Perspectives on Representation and Resolution,
Timm Lichte (Heinrich-Heine-University Düsseldorf, Germany) and Christian Wurm (Heinrich-Heine-University Düsseldorf, Germany)

Week 1, 11:00 – 12:30, Room Mirror Hall, Floor 2

Workshop website: https://ambiguity2018.phil.hhu.de

Natural language is overloaded in the sense that linguistic symbols can, and usually have, two or more (but enumerably many) possible interpretations from which the hearer has to choose a specific one without being explicitely told to do so. This is what we want to call ambiguity, distinguishing it from the notions of underspecification and vagueness. Understood in this general way, ambiguity exists and arises on virtually all levels of linguistic modelling, and resolving ambiguity undoubtly is one of the main challenges when dealing with language in communication.

As a consequence of this diversity, a variety of perspectives exists on how to represent and resolve ambiguity. Also depending on the linguistic object at hand, some represent ambiguity in the semantics by means of, for example, plain disjunction, complex types, or game theoretic models. Some leave it to syntax and assume, for example, two lexical entries for /bank/ that reflect the two different readings. At the same time, however, it is still unclear what semantic ambiguity should be attributed to in logical terms. Finally, if resolution fails, from a philosophical or inferential point of view, one should be interested in the question: given a sentence ambiguous between two readings, does another sentence follow from the two of them? Hence possible treatments of ambiguity include resolution, representation and reasoning. Our main goal is hence to work towards a unified perspective on ambiguity, and for this, we want to bring together researchers from various backgrounds (linguistics, computational linguistics, computer science, logic, philosophy) to approach (among other) the following questions:

  1. How can ambiguity be represented in terms of logic, distributional vectors, weighted (deep) networks etc.?
  2. Is there a “core” notion of ambiguity, and what does it look like? How does it relate to similar phenomena such as polysemy?
  3. What sort of ambiguity should be treated in semantics proper? In particular, when does one want to get rid of ambiguity, and when should we prefer to keep track of it?
  4. What can be gained by combining different approaches to ambiguity, for example simple context-based word disambiguation and disambiguation based on semantic content? Are there underlying methods which always work (e.g. game theory)?
  5. What sort of knowledge is needed for resolving ambiguity? And how can the interaction between knowledge and resolution be modelled?
  6. How can we formalize the distinction between polysemy (including metonymy) and homonymy? Where do we need this distinction, and where not?
  7. What constraints exists on which meanings an ambiguous term can have? And how can we capture this?

Our special focus is thus on bringing together approaches seeing ambiguity as a mere computational problem and approaches seeing it as a linguistic phenomenon with some interest in itself.