UBIU: A Language-Independent System for Coreference Resolution

Proceedings of the 5th International Workshop on Semantic Evaluation, ACL 2010, pages 96–99, Uppsala, Sweden, 15-16 July 2010. c⃝2010 Association for Computational Linguistics UBIU: A Language-Independent System for Coreference Resolution Desislava Zhekova University of Bremen zhekova@uni-bremen.de Sandra K¨ubler Indiana University skuebler@indiana.edu Abstract We present UBIU, a language indepen- dent system for detecting full coreference chains, composed of named entities, pro- nouns, and full noun phrases which makes use of memory based learning and a fea- ture model following Rahman and Ng (2009). UBIU is evaluated on the task “Coreference Resolution in Multiple Lan- guages” (SemEval Task 1 (Recasens et al., 2010)) in the context of the 5th Interna- tional Workshop on Semantic Evaluation. 1 Introduction Coreference resolution is a field in which major progress has been made in the last decade. Af- ter a concentration on rule-based systems (cf. e.g. (Mitkov, 1998; Poesio et al., 2002; Markert and Nissim, 2005)), machine learning methods were embraced (cf. e.g. (Soon et al., 2001; Ng and Cardie, 2002)). However, machine learning based coreference resolution is only possible for a very small number of languages. In order to make such resources available for a wider range of languages, language independent systems are often regarded as a partial solution. To this day, there have been only a few systems reported that work on multiple languages (Mitkov, 1999; Harabagiu and Maio- rano, 2000; Luo and Zitouni, 2005). However, all of those systems were geared towards predefined language sets. In this paper, we present a language indepen- dent system that does require syntactic resources for each language but does not require any effort for adapting the system to a new language, except for minimal effort required to adapt the feature ex- tractor to the new language. The system was com- pletely developed within 4 months, and will be ex- tended to new languages in the future. 2 UBIU: System Structure The UBIU system aims at being a language- independent system in that it uses a combination of machine learning, in the form of memory-based learning (MBL) in the implementation of TiMBL (Daelemans et al., 2007), and language indepen- dent features. MBL uses a similarity metric to find the k nearest neighbors in the training data in order to classify a new example, and it has been shown to work well for NLP problems (Daelemans and van den Bosch, 2005). Similar to the approach by Rahman and Ng (2009), classification in UBUI is based on mention pairs (having been shown to work well for German (Wunsch, 2009)) and uses as features standard types of linguistic annotation that are available for a wide range of languages and are provided by the task. Figure 1 shows an overview of the system. In preprocessing, we slightly change the formatting of the data in order to make it suitable for the next step in which language dependent feature extrac- tion modules are used, from which the training and test sets for the classification are extracted. Our approach is untypical in that it …

Content truncated to 500 words