Semantic Role Induction via Similarity Driven Graph Partitioning

Abstract

In this paper we present a method for unsupervised semantic role
induction which we formalize as a graph partitioning
problem. Argument instances of a verb are represented as vertices in
a graph whose edges express similarities between these instances.
Our graphs consist of multiple edge layers each corresponding to
different ways of operationalizing argument-instance similarity.
Within this general framework, we present two algorithms that differ
in the way they exploit the similarity information encoded in the
graph. The first one is based on agglomeration: two clusters
containing similar instances are grouped into a larger cluster. The
second one is based on propagation: role-label information is
transferred from one cluster to another based on their
similarity. We argue that semantic role induction can be guided by
three linguistic principles and accordingly devise similarity
functions based on them. The first one is the well-known constraint
that semantic roles are unique within a particular frame. The second
one states that the arguments occurring in a specific syntactic
position within a specific linking all bear the same semantic
role. And the third principle suggests that the (asymptotic)
distribution over argument heads is the same for two clusters which
represent the same semantic role.  Experimental results on the
CoNLL~2008 benchmark dataset demonstrate that our approach is able
to infer relatively high-quality shallow semantic representations
whilst delivering results that are competitive with the state of the
art.