<?xml version="1.0" standalone="yes"?>
<Paper uid="W04-2605">
  <Title>Using Selectional Profile Distance to Detect Verb Alternations</Title>
  <Section position="1" start_page="0" end_page="0" type="abstr">
    <SectionTitle>
Abstract
</SectionTitle>
    <Paragraph position="0"> We propose a new method for detecting verb alternations, by comparing the probability distributions over WordNet classes occurring in two potentially alternating argument positions. Existing distance measures compute only the distributional distance, and do not take into account the semantic similarity between Word-Net senses across the distributions. Our method compares two probability distributions over WordNet by measuring the semantic distance of the component nodes, weighted by their probability. To incorporate semantic similarity, we calculate the (dis)similarity between two probability distributions as a weighted distance &amp;quot;travelled&amp;quot; from one to the other through the WordNet hierarchy. We evaluate the measure on the causative alternation, and find that over-all it outperforms existing distance measures.</Paragraph>
  </Section>
class="xml-element"></Paper>