Towards the Annotation of Penn TreeBank with Information Structure

International Joint Conference on Natural Language Processing, pages 1250–1256, Nagoya, Japan, 14-18 October 2013. Towards the Annotation of Penn TreeBank with Information Structure Bernd Bohnet School of Computer Science University of Birmingham Birmingham, UK b.bohnet@cs.bham.ac.uk Alicia Burga DTIC Pompeu Fabra University Barcelona, Spain alicia.burga@upf.edu Leo Wanner ICREA and DTIC Pompeu Fabra University Barcelona, Spain leo.wanner@upf.edu Abstract Information Structure (IS) determines the “communicative” segmentation of the meaning of an utterance, which makes it central to the semantics–syntax– intonation interface and therefore also to NLP. Despite this relevance, IS has not received much attention in the context of the majority of the reference treebanks for data-driven NLP that already contain a semantic and syntactic layers of anno- tation. We present our work in progress on the annotation of the Penn TreeBank with the thematicity dimension of the IS as defined in the Meaning-Text Theory. We experiment with tagging and transition- based parsing techniques. Especially the latter achieve acceptable accuracy with even very small training samples, which is promising for languages with scarce re- sources. 1 Introduction The Information Structure (IS) (aka Topic-Focus Articulation, TFA (Sgall, 1967) in the Prague School and Communicative Structure, CommStr (Mel’ˇcuk, 2001) in the Meaning-Text Theory) de- termines the “communicative” segmentation of the meaning of an utterance. This makes it central to the semantics–syntax–intonation interface (Lam- brecht, 1994; Hajiˇcov´a et al., 1998; Steedman, 2000; Mel’ˇcuk, 2001; Erteschik-Shir, 2007) and therefore also to NLP. However, despite its promi- nence, IS has been largely ignored so far in the context of the reference treebanks for data-driven NLP: Penn Treebank (Marcus et al., 1993) and its semantic counterpart PropBank (Palmer et al., 2005) for English, Tiger (Thielen et al., 1999) for German, Ancora (Taul´e et al., 2008) for Spanish, etc. To the best of our knowledge, only the Prague Dependency Treebank (PDT) (Hajiˇc et al., 2006) is annotated with IS in terms of TFA. This is not to say that no proposals have been made for the an- notation of IS in general; see, e.g., (Calhoun et al., 2005) for English, (Dipper et al., 2004) for Ger- man, (Paggio, 2006) for Danish, etc. However, in the light of the above mentioned interface, it is crucial to have the same corpus annotated with se- mantic, syntactic and IS structures. In this paper, we present our work in progress on the annotation of the ConLL ’09 variant of the Penn TreeBank (PTB) (Hajiˇc et al., 2009) with the thematicity dimension of Mel’ˇcuk’s Comm- Str.1 We have chosen thematicity because (i) it dis- tinguishes, apart from the traditional Theme and Rheme, a Specifier element, and (ii) it is hierarchi- cal in that a thematicity partition can be embedded into another thematicity partition. Both of these features facilitate a fine-grained communicative partition of complex utterances with subordina- tions and thus a more accurate and detailed projec- tion between the different layers of the semantics– syntax–intonation interface. Unlike most other proposals, we aim to au- tomate the annotation procedure and experiment with tagging and transition-based parsing …

Content truncated to 500 words