Carnegie Mellon University
Browse

ParaMor: Minimally-Supervised Induction of Paradigm Structure and Morphological Analysis

journal contribution
posted on 2007-01-01, 00:00 authored by Christian Monson, Jaime G. Carbonell, Alon Lavie, Lori Levin

Paradigms provide an inherent organizational structure to natural language morphology. ParaMor, our minimally supervised morphology induction algorithm, retrusses the word forms of raw text corpora back onto their paradigmatic skeletons; performing on par with state-of-the-art minimally supervised morphology induction algorithms at morphological analysis of English and German. ParaMor consists of two phases. Our algorithm first constructs sets of affixes closely mimicking the paradigms of a language. And with these structures in hand, ParaMor then annotates word forms with morpheme boundaries. To set ParaMor’s few free parameters we analyze a training corpus of Spanish. Without adjusting parameters, we induce the morphological structure of English and German. Adopting the evaluation methodology of Morpho Challenge 2007 (Kurimo et al., 2007), we compare ParaMor’s morphological analyses with Morfessor (Creutz, 2006), a modern minimally supervised morphology induction system. ParaMor consistently achieves competitive F1 measures.

History

Publisher Statement

All Rights Reserved

Date

2007-01-01

Usage metrics

    Exports

    RefWorks
    BibTeX
    Ref. manager
    Endnote
    DataCite
    NLM
    DC