Exploiting cross-linguistic similarities in Zulu and Xhosa computational morphology

Abstract

This paper investigates the possibilities that cross-linguistic similarities and dissimilarities between related languages offer in terms of bootstrapping a morphological analyser. In this case an existing Zulu morphological analyser prototype (ZulMorph) serves as basis for a Xhosa analyser. The investigation is structured around the morphotactics and the morphophonological alternations of the languages involved. Special attention is given to the so-called "open" class, which represents the word root lexicons for specifically nouns and verbs. The acquisition and coverage of these lexicons prove to be crucial for the success of the analysers under development. The bootstrapped morphological analyser is applied to parallel test corpora and the results are discussed. A variety of cross-linguistic effects is illustrated with examples from the corpora. It is found that bootstrapping morphological analysers for languages that exhibit significant structural and lexical similarities may be fruitfully exploited for developing analysers for lesser-resourced languages.

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

“…These features are emblematic for all Bantu languages, and, hence, the results presented in the previous sections are, in its generality, equally applicable to the other Bantu languages. Such transferability at least among two Nguni languages also has been observed in natural language understanding (Pretorius and Bosch 2009), and it may be of interest to check that against the grammar of the Swahili language manager SALAMA 7 , with Swahili being in another Guthrie zone.…”

Strategies for building wordnets for under-resourced languages: The case of African languages

“…The most rudimentary way of doing this was to provide a continually updated list of terms that other languages in the AWN had already included in their wordnets. Languages such as isiXhosa and isiZulu that belong to the same language group show similar morphological and orthographic patterns as demonstrated by Pretorius and Bosch (2009). It therefore makes sense to use the data developed for the related language productively and expand on it, rather than starting new development from the ground up.…”

An assessment of orthographic similarity measures for several African languages

“…For those languages that are 'in-between', it is not just a case of writing the prefixes to a stem disjunctively or together, but only some of the parts of speech or concords are, and then it is likely that for different languages, different choices have been made as to what to write separately and what together. The consequence of this is that, despite a promising case study [15], it is not at all clear whether that particular bootstrapping approach is reusable for other Bantu languages and that instead new rules have to be devised each time. On the positive side, that languages in the lower cluster in Figure 1 are not statistically significantly different does indicate good prospects of reusability of techniques with comparatively little adaptation not just between the three know to be similar languages-isiZulu, isiXhosa, and Ndebele-but also Shona and Runyankore.…”

“…Isolated experiments have been carried out to that extent. For instance, in the knowledgedriven approach, a morphological analyser developed specifically for isiZulu was used to bootstrap one for Ndebele [16] and Setswana [15], a linguistic ontology framework for the noun class system [8], and bootstrapping Runyankore resources from isiZulu [7]. Data-driven (statistical) approaches mainly allude to the hope of transferability across languages [13,20], or obtaining only limited to modest success; e.g., [5], in searching for isiZulu affixes, misses prefixes (e.g., ulu-) and 1/3 of the mere 9 suffixes found were not isiZulu suffixes.…”