Palmyra: A Platform Independent Dependency Annotation Tool for Morphologically Rich Languages

PALMYRA: A Platform Independent Dependency Annotation Tool for Morphologically Rich Languages Talha Javed, Nizar Habash, Dima Taji Computational Approaches to Modeling Language Lab New York University Abu Dhabi {talha.javed, nizar.habash, dima.taji}@nyu.edu Abstract We present PALMYRA, a platform independent graphical dependency tree visualization and editing software. PALMYRA has been specifically designed to support the complexities of syntactic annotation of morphologically rich languages, especially regarding easy change of morphological tokenization through edits, additions, deletions, splits and merges of words. PALMYRA uses an intuitive drag- and-drop interface for editing tree structures, and provides pop-up boxes and keyboard shortcuts for part-of-speech and link label tagging. Keywords: Annotation, Interfaces, Syntax, Morphologically Rich Languages 1. Introduction Producing high-quality natural language syntactic annota- tion is expensive. Well-known large-scale syntactic anno- tation projects, such as the Penn Treebank (Marcus et al., 1993) and the Prague Dependency Treebank (Böhmová et al., 2003), relied on expert linguists to produce carefully an- notated data. This process is rather costly, and as a result, such annotation projects have been undertaken for only a handful of important languages. Efficient annotation tools play an important role in low- ering treebank development costs and enabling the creation of larger, higher quality treebanks. The typical approach is to automatically create syntactic annotations, which are then manually corrected. In this scenario, a goal of the an- notation tool is to lower the annotation burden on the an- notators as much as possible. For morphologically rich languages (MRLs), such as Arabic and Hebrew, syntactic annotation includes morphological ambiguity resolution as well as tokenization adjustment. The change of tokeniza- tion affects the total token count of a sentence and requires adding or deleting tree tokens. For example, the word ����� � wjdhA1 may be automatically analyzed as ��+��� � wjd/VERB+hA/PRON ‘he found her’, but needs to be cor- rected to ��+��� +� w/CONJ+jd/NOUN+hA/PRON2 ‘and her grandfather’ (or vice versa). Since an automatic parser may select an incorrect analysis and tokenization that will need to be corrected, an interface that facilitates such cor- rections during syntactic annotation would be most helpful. PALMYRA is an annotation tool intended to help with the annotation of MRLs in general. However, we focus on Ara- bic here as a representative language for the type of chal- lenges we are interested in addressing. We are primarily in- terested in dependency treebanking and thus the discussion of related work will be focused on dependency representa- tions. 1Arabic transliteration is presented in the Habash-Soudi- Buckwalter scheme (Habash et al., 2007). 2For more information on Arabic morphology and natural lan- guage processing, see (Habash, 2010). 2. Design Aims The following are the design aims for the PALMYRA edi- tor interface. First, we want the editor to provide an easy and direct way to modify the form of a word: by chang- ing it, splitting it, or merging it with other words. Second, we want the editor to allow easy switching between syntac- tic annotation and morphological tokenization to allow the annotators to make these decisions jointly and avoid …

Content truncated to 500 words