Scaling an Irish FST Morphology Engine for Use on Unrestricted Text

This paper details the steps involved in scaling-up a lexicalised finite-state morphology transducer for use on unrestricted text. Our starting point was a base-line inflectional morphology engine [1], with 81% token coverage measured against a 15 million word corpus of Irish texts [2]. Manually scaling the FST lexicon component of a morphology transducer is time-consuming, expensive and rarely, if ever, complete. In order to scale up the engine we used a combination of strategies including semi-automatic population of the finite-state lexicon from machinereadable dictionary resources and from printed resources using optical character recognition, the addition of derivational morphology and the development of morphological guessers. This paper details the coverage increase contributed by each step. The full system achieves token coverage of 93% which is extended to 100% through the use of morphological guessers.