Elliphant: Improved Automatic Detection of Zero Subjects and Impersonal Constructions in Spanish

Proceedings of the 13th Conference of the European Chapter of the Association for Computational Linguistics, pages 706–715, Avignon, France, April 23 - 27 2012. c⃝2012 Association for Computational Linguistics Elliphant: Improved Automatic Detection of Zero Subjects and Impersonal Constructions in Spanish Luz Rello∗ NLP and Web Research Groups Univ. Pompeu Fabra Barcelona, Spain Ricardo Baeza-Yates Yahoo! Research Barcelona, Spain Ruslan Mitkov Research Group in Computational Linguistics Univ. of Wolverhampton, UK Abstract In pro-drop languages, the detection of explicit subjects, zero subjects and non- referential impersonal constructions is cru- cial for anaphora and co-reference resolu- tion. While the identification of explicit and zero subjects has attracted the atten- tion of researchers in the past, the auto- matic identification of impersonal construc- tions in Spanish has not been addressed yet and this work is the first such study. In this paper we present a corpus to under- pin research on the automatic detection of these linguistic phenomena in Spanish and a novel machine learning-based methodol- ogy for their computational treatment. This study also provides an analysis of the fea- tures, discusses performance across two different genres and offers error analysis. The evaluation results show that our system performs better in detecting explicit sub- jects than alternative systems. 1 Introduction Subject ellipsis is the omission of the subject in a sentence. We consider not only missing refer- ential subject (zero subject) as manifestation of ellipsis, but also non-referential impersonal con- structions. Various natural language processing (NLP) tasks benefit from the identification of ellip- tical subjects, primarily anaphora resolution (Mitkov, 2002) and co-reference resolution (Ng and Cardie, 2002). The difficulty in detect- ing missing subjects and non-referential pronouns has been acknowledged since the first studies on ∗ This work was partially funded by a ‘La Caixa’ grant for master students. the computational treatment of anaphora (Hobbs, 1977; Hirst, 1981). However, this task is of cru- cial importance when processing pro-drop lan- guages since subject ellipsis is a pervasive phe- nomenon in these languages (Chomsky, 1981). For instance, in our Spanish corpus, 29% of the subjects are elided. Our method is based on classification of all ex- pressions in subject position, including the recog- nition of Spanish non-referential impersonal con- structions which, to the best of our knowledge, has not yet been addressed. The necessity of iden- tifying such kind of elliptical constructions has been specifically highlighted in work about Span- ish zero pronouns (Ferr´andez and Peral, 2000) and co-reference resolution (Recasens and Hovy, 2009). The main contributions of this study are: • A public annotated corpus in Spanish to compare different strategies for detecting ex- plicit subjects, zero subjects and impersonal constructions. • The first ML based approach to this problem in Spanish and a thorough analysis regarding features, learnability, genre and errors. • The best performing algorithms to automati- cally detect explicit subjects and impersonal constructions in Spanish. The remainder of the paper is organized as fol- lows. Section 2 describes the classes of Spanish subjects, while Section 3 provides a literature re- …

Content truncated to 500 words