Government Domain Named Entity Recognition for South African Languages

Government Domain Named Entity Recognition for South African Languages Roald Eiselen Centre for Text Technology, North-West University, Potchefstroom Campus, South Africa Roald.Eiselen@nwu.ac.za Abstract This paper describes the named entity language resources developed as part of a development project for the South African languages. The development efforts focused on creating protocols and annotated data sets with at least 15,000 annotated named entity tokens for ten of the official South African languages. The description of the protocols and annotated data sets provide an overview of the problems encountered during the annotation of the data sets. Based on these annotated data sets, CRF named entity recognition systems are developed that leverage existing linguistic resources. The newly created named entity recognisers are evaluated, with F-scores of between 0.64 and 0.77, and error analysis is performed to identify possible avenues for improving the quality of the systems. Keywords: Language resource development, South African languages, named entity recognition 1 Introduction Named entity recognition (NER) is the process of automatically classifying different unique identifiers, named entities (NE), according to a predefined set of types. The automatic recognition of named entities is an important, well-understood technology, with a number of scientific events targeting NER (Doddington et al., 2004; Grishman & Sundheim, 1996a, 1996b; Nadeau & Sekine, 2007; Tjong Kim Sang & De Meulder, 2003) and a number of publications on the required data sets, features and algorithms that can be used for the development of NER. Even though extensive work has been done internationally, relatively little work has been done on developing the resources required for creating NE recognisers in the South African languages, apart from Afrikaans (Fourie et al., 2014; Matthew, 2013; Puttkammer, 2006). The research presented here provides an overview of language resources developed for ten South African languages for government domain NER. Named entity recognition can be solved in a number of different ways using rule-based, supervised, semi- supervised or unsupervised methods (Nadeau & Sekine, 2007). Supervised methods typically perform best, but the development of annotated data sets is a time-consuming and training-intensive undertaking, which is why supervised NER systems for most of the South African languages have not been developed to date. This paper describes the process and results of the effort to develop annotated named entity data sets for ten1 of the official languages of South Africa. The first part of the paper describes the data annotation process, with specific focus on the annotation protocols and challenges experienced during the annotation process. The next section provides a short description of the automatic NER systems developed for each language. Finally we provide results for the quality of the NER systems and show F-score results of between 0.64 & 0.77 for the respective languages, with disjunctive languages generally performing better than the conjunctive languages. All of the resources developed as part of this project are available from the Resource Management Agency 2 under the Creative Commons Attribution Licenses. 1 English is excluded since most of these resources are already available for English. 2 Background Over …

Content truncated to 500 words