Semi-automatic Term Extraction for the African Languages, with Special Reference to Northern Sotho

Abstract

Worldwide, semi-automatically extracting terms from corpora is becoming the norm for the compilation of terminology lists, term banks or dictionaries for special purposes. If Africanlanguage terminologists are willing to take their rightful place in the new millennium, they must not only take cognisance of this trend but also be ready to implement the new technology. In this article it is advocated that the best way to do the latter two at this stage, is to opt for computationally straightforward alternatives (i.e. use 'raw corpora') and to make use of widely available software tools (e.g. WordSmith Tools). The main aim is therefore to discover whether or not the semiautomatic extraction of terminology from untagged and unmarked running text by means of basic corpus query software is feasible for the African languages. In order to answer this question a fullblown case study revolving around Northern Sotho linguistic texts is discussed in great detail. The computational results are compared throughout with the outcome of a manual excerption, and vice versa. Attention is given to the concepts 'recall' and 'precision'; different approaches are suggested for the treatment of single-word terms versus multi-word terms; and the various findings are summarised in a Linguistics Terminology lexicon presented as an Appendix.

The Formation of the Image of Top-Ranked Hotels through Real Online Customer Reviews: A Corpus-Based Study of Evaluative Adjectives as Image-Formers/Providers

“…4; It is undeniable that frequency figures are very useful in corpus-based research and that they are normally interpreted in terms of the relevance/importance of the word in the field. Nevertheless, as conceded by Taljard and De Schryver (2002), reading through top-frequency words is obviously an unrefined procedure not necessarily implying significance. This is why, in this paper, the analysis of the image of top-ranked hotels provided by reviewers through EAs will initially be based on the classification of such EAs through a double perspective, based on the frequency but also on the saliency of these EAs.…”

Semi-automatic retrieval of definitional information: a northern Sotho case study

“…Furthermore, access to user-friendly and affordable software such as WordSmith Tools opens the door for terminologists to query and analyse these corpora automatically or at least semi-automatically 2 . It has already been illustrated by Taljard and De Schryver (2002) that it is indeed possible to extract terms semi-automatically from corpora based on subject-field texts, thus reducing (but of course not eliminating) the dependence of the terminologist on the co-operation of the subject-field specialist.…”

“…It has already been pointed out that it is indeed possible for South African terminologists to compile their own special field corpora, and, by following the methodology suggested by Taljard and De Schryver (2002), to semi-automatically extract terms from electronic texts. Should the terminologist now want to add definitions to the extracted terms, he/she is currently left with two options: (a) to formulate definitions with the help of a subject-field expert, or (b) to provide translational equivalents for the terms in English/Afrikaans, then search for definitions in either an LSP dictionary or existing term lists, and as a last step, translate the definitions from English/Afrikaans into the appropriate Bantu language.…”

“…This was the only verb that was found amongst the 50 test terms. In the study done by Taljard and De Schryver (2002), a total number of 350 terms were extracted from the same texts on which the current study is based. Of these, only 14, i.e.…”

Management and Internal Standardization of Chemistry Terminology: A Northern Sotho Case Study

“…However, semi-automatic term extraction does not succeed in extracting all terms from a source text. According to Taljard and De Schryver (2002), semi-automatic term extraction accounts for approximately 60% of terms in a running text. Therefore, computational extraction needs to be complemented by manual term excerption.…”