Compositionality and Distributional Semantic Models
ESSLLI 2010 Workshop

Copenhagen, ESSLLI 2010
August 16-20, 2010
esslli home
In the last ten years distributional semantic models (DSMs), such as LSA, HAL, etc. have been quite successful at addressing semantic similarity, lexical ambiguity, lexical entailment, verb selectional restrictions and other word level relations. In this class of models the meaning of a content word is represented in terms of a distributed vector recording its pattern of cooccurrences (sometimes, in specific syntactic relations) with other content words within a corpus. Different types of semantic tasks and phenomena are then modeled in terms of linear algebra operations on distributional vectors.
A central question about DSMs is whether and how distributional vectors can also be used in the compositional construction of meaning for constituents larger than words, and ultimately for sentences or discourses -- the traditional domains of denotation-based formal semantics. Being able to model key aspects of semantic composition represents a crucial condition for DSMs to provide a more general model of meaning. Conversely, distributional representations might help to model those aspects of meaning that notoriously challenge semantic compositionality, such as semantic context-sensitivity, polysemy, predicate coercion, etc.
The workshop aims to bring together researchers in formal and computational semantics to chart this largely unexplored territory. Some questions to be addressed:
  • Is it possible, and useful, to use Distributional Semantic Models to assign a semantic representation to constituents (e.g. phrases, propositions, etc.)?
  • How can the notion of predication be interpreted in Distributional Semantic Models?
  • Can Distributional Semantic Models provide an alternative way to solve puzzles concerning predicate-argument composition (e.g. type-mismatch, coercion, etc.)?
  • Can we use distributional models to capture argument structure and its alternations, or the Aktionsart of a complex predicates?
  • Can distributional semantic models apply below the word level, characterizing the notions of morpheme productivity and morpheme composition? (e.g. can we capture distributionally the decreasingly compositional meanings of "inter+breed", "inter+act", "inter+view"?)
  • Can distributional semantic models be used to model word meaning interactions in modificational contexts, such as figurative interpretations, context-sensitive sense shifts (e.g. "fast car" vs. "fast guitarist"), etc.?
  • How can polysemy and ambiguity be modelled in distributional semantic models? Which types of ambiguity could be resolved in a DSM-based compositional process? Can this help the task of resolving lexical and textual entailments?
  • What is the right relation between the interpretation functions of formal semantics and the distributional semantic representation these models provide?
  • What should be the most insightful relation between distributional semantic representations of content words and the meaning of the function words that combine with them?
  • Can DSMs provide distributional correlates of constructions and lexical classes that are known to be relevant in formal semantics? (e.g. distributional models of bare plurals, the count vs. mass distinction, generic vs. episodic predicates, etc.).
  • Similarly, can these models capture different types of reference (e.g. nouns or noun phrases that refer to objects, to kinds, to events, to facts or propositions, etc.).

Alessandro Lenci, Department of Linguistics
University of Pisa, Via Santa Maria 36
56126 PISA Italia

Fax: +39 050 2215646
Phone: +39 050 2215638

Roberto Zamparelli Center for Mind/Brain Sciences (CIMeC)
University of Trento, Palazzo Fedrigotti, Corso Bettini, 31
38068 Rovereto (TN), ITALY

Fax: +39 0464 808654
Phone: +39 0464 80 8613

Faculty of Humanities,
						University of Copenhagen