Toward a knowledge-to-text controlled natural language of isiZulu

Abstract

The language isiZulu belongs to the Nguni group of languages, which also include isiXhosa, isiNdebele and siSwati. Of the four Nguni languages, isiZulu is the most dominant language in South Africa, which is spoken by 22.7 % of the country's 51.8 million population. However, isiZulu (and even more so the other Nguni languages) still remains an under-resourced language for software applications. In this article we focus on controlled natural languages for structured knowledge-to-text viewed from a potential utility for verbalising business rules and OWL ontologies. IsiZulu grammar-and by extension, all Bantu languages-shows that a template-based approach is infeasible. This is due to, mainly, the noun class system, the agglutination and verb conjugation with concords for each noun class. We present verbalisation patterns for existential and universal quantification, taxonomic subsumption, axioms with simple properties, and basic cases of negation. Based on the preliminary user assessment of the patterns, selected ones are refined into algorithms for verbalisation to generate correct isiZulu sentences, which have been evaluated.

Evaluation of the E ffects of a S pellchecker on the I ntellectuali s ation of I siZulu

“…Human language technologies for isiZulu are sparse and mostly remain in the realm of theory and academic proof-of-concept tools, such as a morphological analyser (Pretorius and Bosch, 2003), machine translation (Kotzé & Wolff 2015), search engines (Malumba et al 2015), and knowledgeto-text natural language generation (Keet & Khumalo 2017). The main drivers for end-user tools at present are large multinational companies, such as Google Inc. with its rudimentary GoogleTranslate for isiZulu and the localisation efforts of its search engine interface (at no monetary cost), and Microsoft's isiZulu localisation as a for-payment localisation extension/plugin.…”

Evaluation of a Runyankore grammar engine for healthcare messages

“…These systems apply the template NLG technique. However, as was demonstrated in (Keet and Khumalo, 2017), templates are inapplicable to agglutinating Bantu languages, such as Runyankore.…”

Toward an NLG System for Bantu languages: first steps with Runyankore (demo)

“…As for language, we are interested in Runyankore, a Bantu language indigenous to south western Uganda. The highly agglutinative structure and complex verbal morphology of Runyankore make existing NLG systems based on templates inapplicable (Keet and Khumalo, 2017). There have been efforts undertaken to apply the grammar engine technique instead (Byamugisha et al, 2016a;Byamugisha et al, 2016b;Byamugisha et al, 2016c), which resulted in theoretical advances in verbalization rules for ontologies, pluralization of nouns, and verb conjugation that address the text generation needs for Runyankore.…”