Multi-Objective Optimization for the Joint Disambiguation of Nouns and Named Entities

Proceedings of the 53rd Annual Meeting of the Association for Computational Linguistics and the 7th International Joint Conference on Natural Language Processing, pages 596–605, Beijing, China, July 26-31, 2015. c⃝2015 Association for Computational Linguistics Multi-Objective Optimization for the Joint Disambiguation of Nouns and Named Entities Dirk Weissenborn, Leonhard Hennig, Feiyu Xu and Hans Uszkoreit Language Technology Lab, DFKI Alt-Moabit 91c Berlin, Germany {dirk.weissenborn, leonhard.hennig, feiyu, uszkoreit}@dfki.de Abstract In this paper, we present a novel approach to joint word sense disambiguation (WSD) and entity linking (EL) that combines a set of complementary objectives in an exten- sible multi-objective formalism. During disambiguation the system performs con- tinuous optimization to find optimal prob- ability distributions over candidate senses. The performance of our system on nomi- nal WSD as well as EL improves state-of- the-art results on several corpora. These improvements demonstrate the importance of combining complementary objectives in a joint model for robust disambiguation. 1 Introduction The task of automatically assigning the correct meaning to a given word or entity mention in a document is called word sense disambiguation (WSD) (Navigli, 2009) or entity linking (EL) (Bunescu and Pasca, 2006), respectively. Suc- cessful disambiguation requires not only an un- derstanding of the topic or domain a document is dealing with, but also a deep analysis of how an in- dividual word is used within its local context. For example, the meanings of the word “newspaper”, as in the company or the physical product, often cannot be distinguished by the global topic of the document it was mentioned in, but by recogniz- ing which type of meaning fits best into the local context of its mention. On the other hand, for an ambiguous entity mention such as a person name, e.g., “Michael Jordan”, it is important to recognize the domain or topic of the wider context to distin- guish, e.g., between the basketball player and the machine learning expert. The combination of the two most com- monly employed reference knowledge bases for WSD and EL, WordNet (Fellbaum, 1998) and Wikipedia, in BabelNet (Navigli and Ponzetto, 2012), has enabled a new line of research towards the joint disambiguation of words and named en- tities. Babelfy (Moro et al., 2014) has shown the potential of combining these two tasks in a purely knowledge-driven approach that jointly finds connections between potential word senses on a global, document level. On the other hand, typical supervised methods (Zhong and Ng, 2010) trained on sense-annotated datasets are usually quite successful in dealing with individual words in their local context on a sentence level. Hoffart et al. (2011) recognize the importance of combin- ing both local and global context for robust dis- ambiguation. However, their approach is limited to EL and optimization is performed in a discrete setting. We present a system that combines disambigua- tion objectives for both global and local contexts into a single multi-objective function. The result- ing system is flexible and easily extensible with complementary objectives. In contrast to prior work (Hoffart et al., 2011; Moro et …

Content truncated to 500 words