UvT: Memory-Based Pairwise Ranking of Paraphrasing Verbs

Proceedings of the 5th International Workshop on Semantic Evaluation, ACL 2010, pages 260–263, Uppsala, Sweden, 15-16 July 2010. c⃝2010 Association for Computational Linguistics UvT: Memory-based pairwise ranking of paraphrasing verbs Sander Wubben Tilburg centre for Cognition and Communication Tilburg University The Netherlands s.wubben@uvt.nl Abstract In this paper we describe Mephisto, our system for Task 9 of the SemEval-2 work- shop. Our approach to this task is to de- velop a machine learning classifier which determines for each verb pair describing a noun compound which verb should be ranked higher. These classifications are then combined into one ranking. Our clas- sifier uses features from the Google N- gram Corpus, WordNet and the provided training data. 1 Introduction We interpret the task of ranking a set of given paraphrasing verbs as described by Butnariu et al (2010) as a competition between these verbs. Each verb competes with every other verb in the set and receives a positive score if it is more likely to describe the given noun compound (NC) than the other verb and a negative score if it is less likely to describe the NC. In line with this ap- proach we regard the task as a classification prob- lem where for each comparison our classification algorithm picks the paraphrasing verb that is more likely to describe the NC. This brings the clas- sification problem down to three classes: higher, equal or lower. Sometimes the paraphrasing verbs are accompanied by a preposition. In this paper we will simply refer to all verbs and verb-prepositions as verbs. The distribution of the verbs in the training data provides us already with valuable information. We incorporate basic features describing this distribu- tion to train our classifier. We also need addi- tional semantic features that provide us with in- sight into the relation between the NC and the verb, therefore we use features constructed from WordNet and the Google N-gram Corpus to train our Memory-based paraphrase interpretation scor- ing tool (Mephisto). 2 System Description The system consists of three components: the fea- ture extraction component, the classification com- ponent and the ranking component. We will de- scribe all three components. 2.1 Feature Extraction For each verb describing an NC we try to extract those features that describe the probability that this verb is a good interpretation of the NC. We assume that given a NC N1N2 and a verb V , the NC inter- pretation should be N2V N1. The phrase “Butter made from peanuts” adequately describes peanut butter. The training data provides us with a total of 17,727 instances of NC verb pairs scored by hu- man judges. This can be broken down into 4,360 unique verb phrases describing 250 NCs. This distribution already gives us a good clue when we are generating new rankings. The following are the features we used: Weighted mean in training data For each NC that has to be ranked we find the most similar NC in the training data by measuring the overlap in verb phrases between …

Content truncated to 500 words