Distinguishing Common and Proper Nouns

Second Joint Conference on Lexical and Computational Semantics (*SEM), Volume 1: Proceedings of the Main Conference and the Shared Task, pages 80–84, Atlanta, Georgia, June 13-14, 2013. c⃝2013 Association for Computational Linguistics Distinguishing Common and Proper Nouns Judita Preiss and Mark Stevenson {j.preiss, r.m.stevenson}@sheffield.ac.uk Department of Computer Science, University of Sheffield 211 Portobello, Sheffield S1 4DP United Kingdom Abstract We describe a number of techniques for auto- matically deriving lists of common and proper nouns, and show that the distinction between the two can be made automatically using a vector space model learning algorithm. We present a direct evaluation on the British Na- tional Corpus, and application based evalua- tions on Twitter messages and on automatic speech recognition (where the system could be employed to restore case). 1 Introduction Some nouns are homographs (they have the same written form, but different meaning) which can be used to denote either a common or proper noun, for example the word apple in the following examples: (1) Apple designs and creates iPod (2) The Apple II series is a set of 8-bit home computers (3) The apple is the pomaceous fruit of the apple tree (4) For apple enthusiasts – tasting notes and apple identification. The common and proper uses are not always as clearly distinct as in this example; for example, a specific instance of a common noun, e.g., District Court turns court into a proper noun. While heuristically, proper nouns often start with a capital letter in English, capitalization can be in- consistent, incorrect or omitted, and the presence or absence of an article cannot be relied on. The problem of distinguishing between common and proper usages of nouns has not received much attention within language processing, despite being an important component for many tasks including machine translation (Lopez, 2008; Hermjakob et al., 2008), sentiment analysis (Pang and Lee, 2008; Wil- son et al., 2009) and topic tracking (Petrovi´c et al., 2010). Approaches to the problem also have appli- cations to tasks such as web search (Chen et al., 1998; Baeza-Yates and Ribeiro-Neto, 2011), and case restoration (e.g., in automatic speech recogni- tion output) (Baldwin et al., 2009), but frequently involve the manual creation of a gazeteer (a list of proper nouns), which suffer not only from omissions but also often do not allow the listed words to as- sume their common role in text. This paper presents methods for generating lists of nouns that have both common and proper usages (Section 2) and methods for identifying the type of usage (Section 3) which are evaluated using data de- rived automatically from the BNC (Section 4) and on two applications (Section 5). It shows that it is difficult to automatically construct lists of ambigu- ous nouns but also that they can be distinguished ef- fectively using standard features from Word Sense Disambiguation. 2 Generating Lists of Nouns To our knowledge, no comprehensive list of com- mon nouns with proper noun usage is available. We develop a number of heuristics to generate …

Content truncated to 500 words