Neurale netwerke as moontlike woordafkappingstegniek vir Afrikaans

Abstract

In Afrikaans, soos in NederJands en Duits, word saamgestelde woorde aanmekaar geskryf. Nuwe woorde word dus voortdurend gskep deur woorde aanmekaar te haak Dit bemoeilik die proses van woordafkapping tydens teksprosessering, wat deesdae deur rekenaars gedoen word, aangesien die verwysingsbron gedurig verander. Daar bestaan verskeie afkappingsasalgoritmes en tegnieke, maar die resultate is onbevredigend. Afrikaanse woorde met korrekte lettergrecpverdeling is net die elektroniese weergawe van die handwoordeboek van die Afrikaanse Taal (HAT) onttrek. 'n Neutrale netwerk ( vorentoevoer-teru~~;propagering) is met sowat. 5 000 van hierdie woorde afgerig. Die neurale netwerk is verfyn deur 'n gcskikte afrigtingsalgoritme en oorcira.gfunksie vir die probleem asook die optimale aantal verborge lae en aantal neurone in elke laag te bepaal. Die neurale netwerk is met 5 000 nuwe woorde gctoc1.s en dit het 97,5G% van mooutlike posisies korrek as Of geldige of ongeldige aikappingsposisies geklassifiseer. Verder is 510 woorde ui1. tyd<~krifarLikels met die neurale netwerk getoets en 98,75% van moontlike posisie.CJ is korrek geklassifiseer.

In Afrikanus, like in Dutch and German, compound words are written as one word. New words nre thP.refore creat.ed by simply joining words. Word hyphenation during typesetting by computer is a problem, because the source of reference changes all LLe time. Several algorithms and techniques for hyphenation exist, but rcsult.s are not satisfactory. Afrikaans words with correct syllabiftcatiou were extracted from the electronic version of the Ilandwoordeboek van die A/ri.kaansc 1hal (HAT). A neural network (fccoforword backpropagaLion) was trained with about 5 000 of these words. The neural network was refined by heuristically fmding a suitable training algorithm and transfer function for the problem as well as determining the optimal number of layers and number of neurons in ea.ch layer. The neural network was tested with 5 000 words not the training data. It classified 97,56% of possible points in these words correctly as either valid or invalid hyphenation points. Furthermore, f)lO words from articles in a magazine were tested with the neural network and 98,75% of possible positions were classified conectly.