Enriching Morphological Lexica through Unsupervised Derivational Rule Acquisition

Abstract : In a morphological lexicon, each entry combines a lemma with a specific inflection class, often defined by a set of inflection rules. Therefore, such lexica usually give a satisfying account of inflectional operations. Derivational information, however, is usually badly covered. In this paper we introduce a novel approach for enriching morphological lexica with derivational links between entries and with new entries derived from existing ones and attested in large-scale corpora, without relying on prior knowledge of possible derivational processes. To achieve this goal, we adapt the unsupervised morphological rule acquisition tool MorphAcq (Nicolas et al., 2010) in a way allowing it to take into account an existing morphological lexicon developed in the Alexina framework (Sagot, 2010), such as the Lefff for French and the Leffe for Spanish. We apply this tool on large corpora, thus uncovering morphological rules that model derivational operations in these two lexica. We use these rules for generating derivation links between existing entries, as well as for deriving new entries from existing ones and adding those which are best attested in a large corpus. In addition to lexicon development and NLP applications that benefit from rich lexical data, such derivational information will be particularly valuable to linguists who rely on vast amounts of data to describe and analyse these specific morphological phenomena.
Document type :
Conference papers
Liste complète des métadonnées

Cited literature [38 references]  Display  Hide  Download

https://hal.inria.fr/inria-00617064
Contributor : Benoît Sagot <>
Submitted on : Thursday, August 25, 2011 - 7:17:43 PM
Last modification on : Thursday, February 21, 2019 - 12:52:01 PM

File

WoLeR_2011_-_Walther_Nicolas.p...
Files produced by the author(s)

Identifiers

  • HAL Id : inria-00617064, version 1

Citation

Géraldine Walther, Lionel Nicolas. Enriching Morphological Lexica through Unsupervised Derivational Rule Acquisition. WoLeR 2011at ESSLLI : International Workshop on Lexical Resources, Aug 2011, Ljubljana, Slovenia. ⟨inria-00617064⟩

Share

Metrics

Record views

314

Files downloads

313