LATTICE BASED MORPHOLOGICAL RULE INDUCTION
Acta Universitatis Apulensis · 2017
In this paper we will present a novel inflection learning method using lattice based algorithms.The inflection grammar system is represented with a lattice of elementary inflection rules, enabling the description of prefix, infix and suffix transformations as well.The proposed grammar system is ideal for highly agglutinative languages like Hungarian.We created a learning algorithm to generate the rules in an automated way from training word pair sets.Our evaluation shows that this novel method can learn both suffix and infix transformations unlike our baseline algorithm called TASR, and has a better correctness ratio than finite state transducers, too.