Towards Perfection of Machine Learning of Competing Patterns: The Use Case of Czechoslovak Patterns Development

Varování

Publikace nespadá pod Filozofickou fakultu, ale pod Fakultu informatiky. Oficiální stránka publikace je na webu muni.cz.
Autoři

SOJKA Ondřej SOJKA Petr

Rok publikování 2023
Druh Článek ve sborníku
Konference Recent Advances in Slavonic Natural Language Processing (RASLAN 2023)
Fakulta / Pracoviště MU

Fakulta informatiky

Citace
www fulltext PDF
Klíčová slova dictionary problem; effectiveness; hyphenation patterns; patgen; syllabification; Czech; Slovak; Czechoslovak patterns; machine learning
Popis Finding space- and time-effective even perfect solution to the dictionary problem is an important practical and research problem, which solving may lead to a breakthrough in computation. Competing pattern technology from TeX is a special case, where for a given dictionary a word segmentation is stored in the competing patterns yet with very good generalization quality. Recently, the unreasonable effectiveness of pattern generation has been shown---it is possible to use hyphenation patterns to solve the dictionary problem jointly even for several languages without compromise.

In this article, we study the effectiveness of patgen for the supervised machine learning of the generation of the Czechoslovak hyphenation patterns. We show the machine learning techniques to develop competing patterns that are close to being perfect. We evaluate the new approach by improvements and space savings we gained during the development and finetuning of Czechoslovak hyphenation patterns.

Související projekty:

Používáte starou verzi internetového prohlížeče. Doporučujeme aktualizovat Váš prohlížeč na nejnovější verzi.