Section: New Results

Extracting Derivational Relations from an Inflectional Lexicon

Participants : Marion Baranes, Benoît Sagot.

Derivational morphological can provide useful information for natural language processing tasks. Indeed, it can improve any application which has to deal with unknown words such as information extraction, spell-checking and others.

We define a morphological family as a set of semantically related lexical entries which differ by their prefix and/or suffix, thus limiting ourselves to concatenative derivational morphology. We shall denote as derivationally related two morphological lexical entries that belong to the same morphological family.

We have developed a system which performs an analogy-based unsupervized extraction of weighted transformation rules that relate derivationally related lexical entries, and use these rules for extracting derivational relations within an existing inflectional lexicon. Our transformation rules can also be used to infer morphological information (both inflectional and derivational) for wordforms unknown to the inflectional lexicon. Our system is language-independent, although restricted to concatenative derivational morphology. We have evaluated it on four languages: English, French, German and Spanish. Our results will be published at the LREC 2014 conference.