William Lamb

Developing an Automatic Part-of-Speech Tagger for Scottish Gaelic

This paper describes an on-going project that seeks to develop the first automatic PoS tagger for Scottish Gaelic. Adapting the PAROLE tagset for Irish, we manually re-tagged a preexisting 86k token corpus of Scottish Gaelic. A double-verified subset of 13.5k tokens was used to instantiate eight statistical taggers and verify their accuracy, via a randomly assigned hold-out sample. An accuracy level of 76.6% was achieved using a Brill bigram tagger. We provide an overview of the project’s methodology, interim results and future directions.