A Corpus-based Approach to Linguistic Function

A Corpus-Based Approach to Linguistic Function Hengbin Yan and Jonathan Webster The Halliday Centre for Intelligent Applications of Language Studies Department of Chinese, Translation and Linguistics City University of Hong Kong {hbyan2, ctjjw}@cityu.edu.hk Abstract In this paper, we present our recent experience in constructing a first-of-its-kind functional corpus based on the theoretical framework of Systemic Functional Linguistics. Annotated on selected texts from the Penn Treebank, the corpus was built by a collaborative team on web-based annotation platform with several advanced features. After a discussion on the background and motivation of the project, we present our solutions to some of the challenges encountered in the collaborative annotation process. With fine-grained annotations of an initial corpus now available, the corpus can serve as a valuable linguistic resource that complements existing semantically annotated corpora and aid in the development of a larger- scale resource crucial for automated systems for analysis of linguistic function. 1 Introduction Recent years have seen data-driven approaches to natural language processing successfully ap- plied to a wide range of problems including syn- tactic (Collins, 2003; Klein and Manning, 2003) semantic (Gildea and Jurafsky, 2002; Pradhan et al., 2004) and discourse (Hernault et al., 2010) analysis. Computational processing of functional aspects of linguistic data, on the other hand, is a relatively underexplored research area. In lin- guistics, functional analysis refers to the study of language use in context. Among the theories for analyzing the functions of language, Systemic Functional Linguistics (SFL, Halliday and Matthiessen, 2004) is a linguistic framework that is becoming increasing influential in recent years. SFL provides an ideal handle to exploring lan- guage as intentional acts of meaning, comple- menting more syntactically oriented approaches to linguistic study. Despite its power, traditional analysis with SFL is done manually, a time- and effort- consuming process. We are motivated in our study to extend the power of the framework to computational analy- sis. The difficulty in automating analysis of lin- guistic functions lies in both the fuzziness in the functional domain and a lack of relevant compu- tational resources. The most significant lack of resource is a high-quality reference corpus cru- cial to statistical analysis and modeling. In the following sections, we discuss our initial efforts in constructing such a resource on a collaborative annotation platform and present the initial results from the corpus. The corpus is our first step in bridging the gap between the linguistic theory and application of such theory including auto- mated analysis of language functions. 2 Related Works Over the past decades, the construction of prom- inent linguistic corpora to account for the syntac- tic (Marcus, 1993), semantic (Kingsbury et al., 2002) and discourse (Carlson and Okurowski, 2002; Prasad et al., 2008) structures of linguistic information has deepened our understanding in each layer and made possible automated data- driven analysis based on them. Although the ad- vantages of a functional-semantic orientation are apparent to text analysis, the complexity arising from annotation of multi-level functional- semantic information, such as that found in SFL, has led to …

Content truncated to 500 words