South African Language Resources: Phrase Chunking

South African Language Resources: Phrase Chunking Roald Eiselen Centre for Text Technology, North-West University, Potchefstroom Campus, South Africa Roald.Eiselen@nwu.ac.za Abstract Phrase chunking remains an important natural language processing (NLP) technique for intermediate syntactic processing. This paper describes the development of protocols, annotated phrase chunking data sets and automatic phrase chunkers for ten South African languages. Various problems with adapting the existing annotation protocols of English are discussed as well as an overview of the annotated datasets. Based on the annotated sets, CRF-based phrase chunkers are created and tested with a combination of different features, including part of speech tags and character n-grams. The results of the phrase chunking evaluation show that disjunctively written languages can achieve notably better results for phrase chunking with a limited data set than conjunctive languages, but that the addition of character n-grams improve the results for conjunctive languages. Keywords: Language resource development, South African languages, phrase chunking, shallow parsing 1 Introduction Syntactic analysis is an important part of many natural language processing systems, but generally requires large annotated data sets that are syntactically parsed and disambiguated in order to be effective. Abney (1992) proposed an alternative method for performing partial syntactic analysis in environments where full syntactic analysis is not required, but a level of analysis beyond part of speech is required. Phrase chunking, or shallow parsing, provides a flat representation of the major syntactic categories by assigning tokens to non-recursive segments, without resolving more complex attachment (Abney, 1992; Kübler et al., 2010). This approach allows for syntactic analysis that is useful, but does not require full grammars or large training sets that are complex and expensive to develop. This approach to syntactic analysis is especially interesting in the South African context where work on automatic syntactic analysis has been limited, with little data widely available for research or development purposes. Furthermore, at least nine of the official South African languages have limited syntactic or annotated data resources. An intermediate approach that will allow for some level of syntactic analysis and can be used other NLP technologies and applications, is ideal for this situation. The research described in this paper gives an overview of the NCHLT Text Phase II project, which undertook the task of annotating 15,000 tokens with their phrasal constituency and creating phrase chunkers for ten of the official South African languages. This paper describes the development process for these resources by presenting an overview of some of the challenges and solutions experienced during the development. The first part of the paper presents an overview of the annotated datasets, including the development of protocols for use during the annotation process. Some of the language specific issues and considerations encountered during the annotation process are also discussed. This is followed by a short description of the automatic phrase chunkers that were developed using conditional random fields (CRF). After these descriptions, the phrase chunkers are evaluated and we provide results showing that high quality CRF phrase chunkers for the disjunctive South African languages can be …

Content truncated to 500 words