Correcting Errors in a Treebank Based on Tree Mining

Correcting Errors in a Treebank Based on Tree Mining Kanta Suzuki1, Yoshihide Kato2 and Shigeki Matsubara1 1Graduate School of Information Science, Nagoya University 2Information & Communications, Nagoya University ksuzuki@db.ss.is.nagoya-u.ac.jp Abstract This paper provides a new method to correct annotation errors in a treebank. The previous error correction method constructs a pseudo parallel corpus where incorrect partial parse trees are paired with correct ones, and extracts error correction rules from the parallel corpus. By applying these rules to a treebank, the method corrects errors. However, this method does not achieve wide coverage of error correction. To achieve wide coverage, our method adopts a different approach. In our method, we consider that an infrequent pattern which can be transformed to a frequent one is an annotation error pattern. Based on a tree mining technique, our method seeks such infrequent tree patterns, and constructs error correction rules each of which consists of an infrequent pattern and a corresponding frequent pattern. We conducted an experiment using the Penn Treebank. We obtained 1,987 rules which are not constructed by the previous method, and the rules achieved good precision. Keywords: error correction, synchronous tree substitution grammar, FREQT 1. Introduction It is inevitable for annotated corpora to contain errors caused by manual or semi-manual annotation process. So, detecting and correcting errors in annotated corpora are im- portant tasks. Many studies suggest methods of detecting or correcting errors in various kinds of annotated corpora (see (Dickinson, 2015) for a survey). There are several meth- ods of detecting annotation errors in a phrase structure tree- bank (Dickinson and Meurers, 2003; Ule and Simov, 2004; Dickinson and Meurers, 2005; Boyd et al., 2007; Dickin- son, 2009; Przepi´orkowski and Lenart, 2012; Kulick et al., 2013; Faria, 2014). However, there is little work on tree- bank error correction. One exception is the work of Kato and Matsubara (2010). Their method constructs a pseudo parallel corpus where incorrect parse trees are paired with correct ones, and ex- tracts error correction rules from the parallel corpus. The rules transform incorrect tree patterns to correct ones. By applying these rules to a treebank, the method corrects er- rors. However, this method does not achieve wide coverage of error correction. To solve this problem, we propose another approach to con- struct error correction rules. Our method does not construct a pseudo parallel corpus. In our method, we consider that an infrequent tree pattern which can be transformed to a frequent one is an annotation error pattern. Based on a tree mining technique, our method seeks such infrequent pat- terns efficiently. The method constructs error correction rules by pairing the infrequent tree patterns with the fre- quent ones. We conducted an experiment using the Penn Treebank. We obtained 1,987 rules which are not con- structed by the previous method, and the rules achieved good precision. This paper is organized as follows: Section 2 introduces the previous method of correcting errors in a treebank. Section 3 explains our method which is based on tree mining. Sec- tion 4 …

Content truncated to 500 words