Adapting WSJ-trained parsers to the British National Corpus using in-domain self-training

Jennifer Foster (1) , Joachim Wagner (1) , Djamé Seddah (2) , Josef van Genabith (1)

1 NCLT - National Centre for Language Technology
2 LaLIC - Langues, logiques, informatiques, cognition

We introduce a set of 1,000 gold standard parse trees for the British National Corpus (BNC) and perform a series of self-training experiments with Charniak and Johnson's reranking parser and BNC sentences. We show that retraining this parser with a combination of one million BNC parse trees (produced by the same parser) and the original WSJ training data yields improvements of 0.4% on WSJ Section 23 and 1.7% on the new BNC gold standard set.

Mots clés: parsing; Probabilistic parsing; out of domain parsing; BNC

Domaines: Traitement du texte et du document

Jennifer Foster, Joachim Wagner, Djamé Seddah, Josef van Genabith. Adapting WSJ-trained parsers to the British National Corpus using in-domain self-training. Proceedings of the 10th International Conference on Parsing Technologies : IWPT '07, Association for Computational Linguistics, 2007, Prague, Czech Republic. pp.33--35. ⟨inria-00545429⟩