A Three-Stage Morphological Pipeline for Word Form Analysis in Yakut
Nicolas Cortegoso Vissio, Maria V. Khokhlova · 2024
Rule-based morphological analyzers for agglutinative languages typically rely on a lexicon of roots and affixes, along with a set of morphophonological rules to analyze and generate word forms. While such analyzers achieve high precision, they often suffer from limited recall, as they cannot process word forms with roots that are not explicitly included in the lexicon. Since it is impossible to account for all potential roots in a language, handling unknown words remains a significant obstacle in rule-based morphological analysis. This paper addresses this challenge using the Yakut language as a case study. We propose a three-stage morphological processing pipeline, where specialized transducers are applied sequentially when the previous one fails to analyze a word form. At each stage of the pipeline, precision is partially sacrificed to improve recall. The goal is to strike a balance that maximizes coverage while minimizing ambiguity in the resulting analyses.