Calculating Semantic Distance between Word Sense Probability Distributions

Calculating Semantic Distance between Word Sense Probability Distributions Vivian Tsang and Suzanne Stevenson Department of Computer Science University of Toronto � vyctsang,suzanne � @cs.toronto.edu Abstract Semantic similarity measures have focused on individual word senses. However, in many ap- plications, it may be informative to compare the overall sense distributions for two differ- ent contexts. We propose a new method for comparing two probability distributions over WordNet, which captures in a single measure the aggregate semantic distance of the com- ponent nodes, weighted by their probability. Previous such measures compute only the dis- tributional distance, and do not take into ac- count the semantic similarity between Word- Net senses across the distributions. To in- corporate semantic similarity, we calculate the (dis)similarity between two probability distri- butions as a weighted distance “travelled” from one to the other through the WordNet hierar- chy. We evaluate the measure by applying it to the acquisition of verb argument alternation knowledge, and find that overall it outperforms existing distance measures. 1 Introduction Much attention has recently been given to calculating the similarity of word senses, in support of various natu- ral language learning and processing tasks. Such tech- niques apply within a semantic hierarchy, or ontology, such as WordNet. Typical methods comprise an edge- distance measurement over the two sense nodes being compared within the hierarchy (Leacock and Chodorow, 1998; Rada et al., 1989; Wu and Palmer, 1994). Other approaches instead assume a probability distribution over the entire sense hierarchy; similarity is captured between individual senses by a formula over the information con- tent (negative log probabilities) of relevant nodes (e.g., Jiang and Conrath, 1997; Lin, 1998). The latter case assumes that there is a single WordNet probability distribution of interest, which is estimated by populating the hierarchy with word frequencies from an appropriate corpus (e.g, Jiang and Conrath, 1997). But some problems more naturally give rise to multiple con- ditional probability distributions estimated from counts that are conditioned on various contexts, such as differ- ent corpora or differing word usage within a single cor- pus. Each of these contexts would yield a distinct Word- Net probability distribution, or what we will call a sense profile. In this situation, instead of asking how similar are two senses within a single sense profile, one may want to know how similar are two sense profiles—i.e., two (con- ditional) distributions across the entire set of nodes. This question could be important to a number of ap- plications. When two sets of WordNet frequency counts are conditioned on differing contexts, a comparison of the resulting probability distributions can give us a measure of the degree of semantic similarity of the conditioning contexts themselves. These conditioning contexts may be any relevant ones defined by the application, such as dif- fering sets of documents (to support asking how similar various document collections are), or differing usages of words within or across document collections (to support asking questions about the similarity of various words in their usages). For example, we foresee comparing the …

Content truncated to 500 words