Applying automatically parsed corpora to the study of language variation

Proceedings of COLING 2014, the 25th International Conference on Computational Linguistics: Technical Papers, pages 1974–1984, Dublin, Ireland, August 23-29 2014. Applying automatically parsed corpora to the study of language variation Jelke Bloem Arjen Versloot Amsterdam Center for Language and Communication University of Amsterdam 1012 VB Amsterdam, Netherlands {j.bloem, a.p.versloot, f.p.weerman}@uva.nl Fred Weerman Abstract In this work, we discuss the benefits of using automatically parsed corpora to study language variation. The study of language variation is an area of linguistics in which quantitative methods have been particu- larly successful. We argue that the large datasets that can be obtained using automatic annotation can help drive further research in this direction, providing sufficient data for the increasingly complex models used to describe variation. We demonstrate this by replicating and extending a previous quantitative variation study that used manually and semi-automatically annotated data. We show that while the study cannot be replicated completely due to limitations of the existing automatic annotation, we can draw at least the same conclusions as the original study. In addition, we demonstrate the flexibility of this method by extending the findings to related linguistic constructions and to another domain of text, using additional data. 1 Introduction There are many examples of linguistic variation that are not easily explained in terms of rules. One may find two grammatically correct constructions that can be used to express similar meanings, yet speakers still use both of them. A well-known example in English is the dative alternation, where a transitive verb such as give can be phrased as a double object construction (1) or as a prepositional dative (2): (1) He gave his friend the ticket. (2) He gave the ticket to his friend. Studies of such phenomena tend to find that there are multiple variables that may influence whether a speaker chooses one or the other construction. This has prompted various multivariate studies by quantitative linguists to analyze instances of such variation in language corpora, starting with Gries (2001), or Bresnan et al. (2007) for a study on the dative alternation. The multivariate statistical models that these studies employ can quantify the contribution of each variable to the variation in probabilistic terms, rather than examining them in isolation. We show the benefits of using automatically annotated corpora for the study of language variation by replicat- ing a previous, manual multivariate study on Dutch verbal cluster variation (De Sutter, 2009), and extending it to fit more types of clusters. We also show that the same variables are explanatory in at least two different text domains, Wikipedia text and European Parliament proceedings. The larger scale of our investigation allows us to generalize the claims of the previous study to Dutch two-verb cluster variation in general. This topic makes for a good methodological case study for the use of automatically annotated corpora, as verbal clusters are a widely studied phenomena in Dutch syntactic literature (Evers, 1975; Den Besten and Edmondson, 1983; Haege- man and van Riemsdijk, 1986; Zwart, 1996; Wurmbrand, 2004), and the optionality …

Content truncated to 500 words