Converting the Penn Treebank to Systemic Functional Grammar

Converting the Penn Treebank to Systemic Functional Grammar Matthew Honnibal Department of Linguistics, Macquarie University Macquarie University 2109 Sydney Australia mhonn@it.usyd.edu.au Abstract Systemic functional linguistics offers a grammar that is semantically organised, so that salient gram- matical choices are made explicit. This paper de- scribes the explication of these choices through the conversion of the Penn Treebank into a systemic functional grammar corpus. Developing such a re- source can help connect work in natural language processing to a significant body of research dealing explicitly with the issue of how lexical and gram- matical selections create meaning. 1 Introduction The Penn Treebank was designed to maximise con- sistency and annotator efficiency, rather than con- formity with any particular linguistic theory (Mar- cus et al., 1994). This results in trees that strongly suggest the use of synthetic features to explicate semantically significant grammatical choices like mood, tense, voice or negation. These distinctions lie latent in the configuration of the tree in the Tree- bank II annotation scheme, making it difficult for a machine learner to make use of them. Rather than the ad hoc addition of this informa- tion at the feature extraction stage, the corpus can be re-presented in a way that makes feature extraction more principled. This involves increasing the size and complexity of the representation of a sentence by organising the tree semantically. Organising a grammar semantically is by no means a trivial task, and has been an active area of linguistic research for the last forty years. This paper describes the con- version of the Penn Treebank into a prominent out- put of such research, systemic functional grammar (SFG). Systemic functional grammar does not confine its description to syntactic structure, but includes a rep- resentation of the choices grammatical configura- tions represent — or ‘realise’, to use the term pre- ferred in the linguistics literature (Halliday, 1976). There is growing evidence that systemic func- tional grammar can be usefully applied to natural language processing (Munro, 2003; Couchman and Whitelaw, 2003), and there is a strong history of interaction between systemic functional linguistics and natural language generation (Matthiessen and Bateman, 1991). However, there is currently a lack of computational SFG resources. There is no stan- dard format for machine readable annotation, no an- notated corpora, and no useable parsers. Converting the Penn Treebank will make a large body of SFG annotated data available to computational linguists for the first time, an important step towards address- ing this situation. We first discuss some preliminaries relating to the nature of systemic functional grammar, and the scope of the converted corpus’s annotation. We then discuss the conversion of the treebank’s phrase- structure representation to SFG constituency struc- ture, and finally we discuss the addition of interper- sonal and textual function structures. 2 Some preliminaries 2.1 Structure of the SFG analysis Systemic functional grammar divides the task of grammatical analysis — the process of stating the grammatical properties of a text — into two parts: analysis of syntactic structures, and analysis of function …

Content truncated to 500 words