Web-based Annotation Tool for Inflectional Language Resources

Web-based Annotation Tool for Inflectional Language Resources Abdulrahman Alosaimy, Eric Atwell School of Computing, University of Leeds Leeds LS2 9JT, UK {scama, e.s.atwell}@leeds.ac.uk Abstract We present Wasim, a web-based tool for semi-automatic morphosyntactic annotation of inflectional languages resources. The tool features high flexibility in segmenting tokens, editing, diacritizing, and labelling tokens and segments. Text annotation of highly inflectional languages (including Arabic) requires key functionality which we could not see in a survey of existing tools. Wasim integrates with morphological analysers to speed up the annotation process by selecting one from their proposed analyses. It integrates as well with external POS taggers for kick-start annotation and adaptive predicting based on annotations made so far. It aims to speed up the annotation by completely relying on a keyboard interface, with no mouse interaction required. Wasim has been tested on four case studies and these features proved to be useful. The source-code is released under the MIT license. Keywords: annotation, inflectional language, morphosyntactic, Arabic 1. Introduction POS tagging text in inflectional languages is usually hard. A typical problem is substantial lexical data sparseness due to the high number of possible inflexions of a single word. To reduce sparseness and number of Out-of- Vocabulary (OOV) words, inflected words are often segmented prior to or in parallel with POS tagging. However, the segmentation process is prone to errors. Inflection boundaries are often not marked which increases the number of homographs (two or more words spelt in the same form but with different POS tag or pronunciation (e.g. due to differences in diacritization). Some orthographical changes are caused by inflexions, making it hard to recover the original word form. As a result, a segmentation process sometimes fails to detect morphemes. Wasim is a web-based tool for semi-automatic annotation of text for the purpose of gold standard corpus production1. It was developed for the annotation of our Sunnah Arabic Corpus (SAC) (Alosaimy & Atwell, 2017), a collection of classical Arabic sayings ascribed to the prophet Mohammad. It has also been tested in four case studies. For the project, we investigated the required set of features needed for annotating SAC and used these as criteria in a survey of existing tools. In our search for currently available tools, we limited our survey to tools that 1. are web-based: to integrate it with other systems, and to allow easier access through browsers. 2. Annotate text tokens with morpho-syntactic tags in CoNLL-U v.2 format (Nivre & Agic, 2017)2. 3. Support right-to-left languages. 4. are available to download for research purposes. Morphosyntactic annotation of SAC (and other highly inflectional language corpora) requires additional specialized functionality: 1 Source code and demo is available at: http://wasim.al- osaimy.com 2 CoNLL-U format has been used in Universal Dependencies project and well described in http://universaldependencies.org/ 1. Segmentation of one word into a set of segments 2. Addition of orthographical accents or diacritics 3. Listing a set of solutions from a lexicon dictionary (internally or externally using a morphological analyser) 4. Consistency validation and integrating annotation guidelines …

Content truncated to 500 words